chore: version packages

- Add HTML sanitizer for markdown rendering (XSS prevention)
- Switch service worker to network-first caching (deploys take effect immediately)
- Sanitize Content-Disposition filenames (header injection prevention)
- Expose session.muxName getter, replace unsafe `as any` cast
- Static import for execFile, update CLAUDE.md keyboard shortcuts

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -11,10 +11,10 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
 
       - name: Setup Node.js
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@v6
         with:
           node-version: 22
           cache: 'npm'

.github/workflows/release.yml

@@ -16,12 +16,12 @@ jobs:
       pull-requests: write
     steps:
       - name: Checkout repo
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Setup Node.js
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@v6
         with:
-          node-version: 20
+          node-version: 22
           cache: npm
           registry-url: https://registry.npmjs.org
 
@@ -42,3 +42,25 @@ jobs:
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+
+      - name: Rename release tag to codeman
+        if: steps.changesets.outputs.published == 'true'
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          VERSION=$(node -p "require('./package.json').version")
+          OLD_TAG="aicodeman@${VERSION}"
+          NEW_TAG="codeman@${VERSION}"
+
+          # Update the GitHub release BEFORE deleting the old tag
+          RELEASE_ID=$(gh release view "$OLD_TAG" --json databaseId -q .databaseId 2>/dev/null || true)
+          if [ -n "$RELEASE_ID" ]; then
+            gh api -X PATCH "repos/${{ github.repository }}/releases/${RELEASE_ID}" \
+              -f tag_name="$NEW_TAG" \
+              -f name="$NEW_TAG"
+          fi
+
+          # Retag
+          git tag "$NEW_TAG" "$OLD_TAG" 2>/dev/null || true
+          git tag -d "$OLD_TAG" 2>/dev/null || true
+          git push origin "$NEW_TAG" ":refs/tags/$OLD_TAG" 2>/dev/null || true

.gitignore

@@ -48,7 +48,12 @@ Thumbs.db
 # Generated output
 out/
 screenshots-echo-diag/
-tools/remotion/out/
+scripts/remotion/out/
+
+# Artifacts that should not be tracked
+test-results/
+tmp/
+public
 
 # Claude Code plan tracking
 plan.json
@@ -60,3 +65,4 @@ media-assets/
 commands
 todo.md
 @fix_plan.md
+readme-preview.mjs

.prettierignore

@@ -6,4 +6,4 @@ src/web/public/app.js
 src/web/public/styles.css
 src/web/public/mobile.css
 src/web/public/index.html
-tools/
+scripts/remotion/

CHANGELOG.md

@@ -1,5 +1,322 @@
 # aicodeman
 
+## 0.5.11
+
+### Patch Changes
+
+- Community contributions and security hardening:
+  - Mobile response viewer: native-scroll panel for reading full Claude responses with markdown rendering via marked.js (PR #62)
+  - PWA support: service worker caching, web app manifest, and Android home screen install (PR #59)
+  - Named Cloudflare tunnel support (PR #58)
+  - Markdown rendering for response viewer with HTML sanitization (XSS prevention) — strips dangerous elements, event handlers, and javascript: URIs
+  - Service worker switched from stale-while-revalidate to network-first caching so deploys take effect immediately
+  - Content-Disposition filename sanitization to prevent header injection in file downloads
+  - Expose session.muxName public getter, replace unsafe `as any` cast in session-routes
+  - Static import for execFile in session-routes
+  - Keyboard shortcut updates: Alt+1-9 tab switching, Shift+Enter newline
+  - Repo restructure for cleaner GitHub landing page
+  - Mobile logo, expandable history, session resume fixes
+
+## 0.5.10
+
+### Patch Changes
+
+- fix: allow bracket characters in model validation regex so models like opus[1m] (1M context window) are accepted instead of silently dropped. Quote the model flag value in tmux spawn commands to prevent bash glob expansion of bracket patterns.
+
+  docs: update macOS launchd instructions to use `launchctl bootstrap` instead of deprecated `load`. Clean up README install and service sections.
+
+## 0.5.9
+
+### Patch Changes
+
+- Mobile keyboard accessory bar: add configurable "Extended Keyboard Bar" setting (Settings > Display > Input) that toggles between simple mode (up/down arrows, /init, /clear, /compact, paste, dismiss) and extended mode (adds left/right arrows, Tab, Shift+Tab, Ctrl+O, Alt+Enter, Esc). Default is simple mode. Setting is device-specific (not synced to server).
+
+  Restyle dismiss button: muted steel-blue tone, fills remaining bar space via flex, larger tap target. Arrow buttons now blue.
+
+  Fix paste overlay visibility on mobile: dialog repositioned to top of screen (15vh from top) so the virtual keyboard doesn't cover it. Textarea enlarged for better usability.
+
+  (Also includes all v0.5.8 changes: case reorder/delete, XSS sanitization, auto-attach PTY on restart, mobile keyboard buttons, macOS installer fixes, terminal flicker fix, state store collision fix.)
+
+## 0.5.8
+
+### Patch Changes
+
+- Case management: add Manage tab with reorder (up/down arrows) and delete for cases; linked cases are unlinked (folder preserved), CASES_DIR cases are permanently deleted. New endpoints: DELETE /api/cases/:name, PUT /api/cases/order. SSE events: case:deleted, case:order-changed.
+
+  Security: sanitize case names from filesystem with /^[a-zA-Z0-9_-]+$/ regex before returning from GET /api/cases to prevent XSS via maliciously-named directories reaching frontend inline onclick handlers.
+
+  Auto-attach PTY: server now calls startInteractive() for recovered tmux sessions during startup so all sessions resume capturing output immediately after deploy, instead of waiting for client selection. Frontend auto-attach condition relaxed from (pid===null && status==='idle') to (pid===null && !\_ended).
+
+  Mobile keyboard accessory: add Shift+Tab, Tab, Esc, Alt+Enter, Left/Right arrow, and Ctrl+O buttons.
+
+  Terminal: fix flicker regression by moving viewport clear inside dimension guard.
+
+  State store: fix temp file collisions on concurrent writes.
+
+  macOS: fix installer failures when piped via curl | bash, add HTML cache support, launchd service template, and trust dialog handling.
+
+  Housekeeping: remove accidentally committed dist/state-store.js build artifact.
+
+## 0.5.7
+
+### Patch Changes
+
+- feat: support "Default (CLI default)" option for model selection. Adds a new empty-value option to the model dropdown that defers to the CLI's own default model instead of forcing a specific model. Ensures empty defaultModel values are treated as undefined when passed to session creation and Ralph loop start, preventing empty strings from being sent as model flags.
+
+## 0.5.6
+
+### Patch Changes
+
+- fix: default new sessions to opus[1m] (1M context window) instead of plain opus (200k context)
+
+## 0.5.5
+
+### Patch Changes
+
+- Add 1M Opus context quick setting — per-case and global toggle that writes `model: "opus[1m]"` to `.claude/settings.local.json` when creating new sessions. Fix mobile layout: banners (respawn, timer, orchestrator) between header and main content now visible by switching from margin-top on `.main` to padding-top on `.app`. Add tablet-optimized respawn banner styles and mobile phone banner refinements.
+
+## 0.5.4
+
+### Patch Changes
+
+- Fix terminal flicker regression — re-add server-side DEC 2026 synchronized output wrapping around batched terminal data. Ink spinner frames (cursor-up + redraw cycles) do not emit their own DEC 2026 markers, so without the server wrapper each partial cursor update rendered individually causing visible flicker. Also: extract SSE stream management, session listener wiring, and respawn event wiring from server.ts into dedicated modules; deduplicate error message extraction across 7 files with shared getErrorMessage() helper; update SSE event count in CLAUDE.md (106 → 117).
+
+## 0.5.3
+
+### Patch Changes
+
+- Readability refactor across 12 core files, extracting ~35 helper methods to reduce duplication:
+  - state-store: extract serializeState(), split assembleStateJson() into focused sub-methods
+  - session: extract \_resetBuffers() (3x dedup), \_clearAllTimers() (10 timer cleanups), \_handleJsonMessage()
+  - ralph-tracker: extract completeAllTodos() (4x dedup), emitValidationWarning(), named similarity constants
+  - subagent-watcher: extract markSubagentAsCompleted(), extractFirstTextContent(), emitToolResult(), findOldestInactiveAgent()
+  - respawn-controller: extract recoveryResetToWatching(), canAutoAccept(), formatRemainingSeconds(), validatePositiveTimeout()
+  - tmux-manager: replace 15 path.includes() with UNSAFE_PATH_CHARS regex, extract buildEnvExports/buildPathExport/\_configureOpenCode helpers
+  - session-auto-ops: extract executeWhenIdle() shared retry helper, convert to options object, add validateThreshold()
+  - app.js: add \_clearTimer() (11 call sites), \_isStaleSelect(), keyboard shortcut lookup table, \_cleanupPreviousSession(), \_resetAllAppState()
+  - route-helpers: add readJsonConfig() (5 inline patterns replaced), validateSessionFilePath() (2 duplicated blocks replaced)
+
+## 0.5.2
+
+### Patch Changes
+
+- Make buffer size limits configurable via CODEMAN\_\* environment variables (MAX_TERMINAL_BUFFER, TRIM_TERMINAL_TO, MAX_TEXT_OUTPUT, TRIM_TEXT_TO, MAX_MESSAGES), falling back to existing defaults. Allows users with fewer sessions or more RAM to tune buffer sizes without patching source.
+
+  Fix duplicate terminal output on tab switch to busy sessions by clearing the terminal before writing the new buffer.
+
+  Fix stale Ink CUP frames after tab switch by sending Ctrl+L to force a clean redraw.
+
+  Fix mobile CJK input handling: resolve textarea positioning, terminal flicker during composition, and layout overflow on small screens. Improve CJK composition lifecycle with better event handling and fallback flush timers.
+
+## 0.5.1
+
+### Patch Changes
+
+- refactor: codebase cleanup — extract route helpers, eliminate boilerplate, optimize hot paths
+  - Add `parseBody()` helper to route-helpers.ts: validates request body against Zod schema with structured 400 error on failure, replacing 37 identical safeParse + error-check blocks across 10 route files
+  - Add `persistAndBroadcastSession()` helper: combines persist + SessionUpdated broadcast into one call, replacing 5 repeated 2-line pairs
+  - Migrate session-routes.ts to use `findSessionOrFail()` consistently (17 inline session lookups replaced) and `parseBody()` (12 patterns)
+  - Migrate ralph-routes.ts to use `findSessionOrFail()` (9 lookups) and `parseBody()` (4 patterns)
+  - Migrate 8 remaining route files to use `parseBody()` (21 patterns total)
+  - Fix O(n log n) eviction in bash-tool-parser.ts: replace `Array.from().sort()[0]` with O(n) min-scan for oldest active tool
+  - Extract `_debouncedCall()` utility in frontend: replaces 4 manual debounce patterns (7 lines each → 1 line) in app.js, panels-ui.js, ralph-panel.js
+  - Net reduction: 208 lines removed across 16 files
+
+## 0.5.0
+
+### Minor Changes
+
+- Visual redesign with glass morphism, refined colors, and polished UI. Optimize history endpoint with buffer reuse and line iterator. Fix Ink frame search window (4KB→64KB) to prevent partial frames. Fix stale terminal data on tab switch via chunkedTerminalWrite cancellation. Improve history prompt extraction with expanded command filtering and tail scan fallback. Align case select group height to match dropdown. Fix no-control-regex lint error for ANSI strip pattern. Add browser-testing-guide to CLAUDE.md references.
+
+## 0.4.7
+
+### Patch Changes
+
+- feat: improve session navigability in history and monitor panel (closes #45)
+  - History items now show the first user prompt as the title with the project path as a subtitle, making it much easier to distinguish sessions from the same project
+  - The `/api/history/sessions` endpoint extracts the first user message from each transcript JSONL, stripping system-injected XML tags and command artifacts, truncating to 120 chars
+  - Monitor panel session rows are now clickable — clicking navigates directly to that session's tab via `selectSession()`; Kill button retains independent behavior via `stopPropagation()`
+  - Updated CLAUDE.md architecture tables to reflect Orchestrator Loop additions (14 route modules, 15 type files, orchestrator domain files, orchestrator-panel.js frontend module)
+  - fix: stop subagent monitor windows from auto-opening on discovery
+  - feat: add Orchestrator Loop with phased plan execution, live progress during plan generation, and toolbar button (hidden until fully tested)
+  - fix: patch 3 production bugs found during deep audit
+  - fix: restore mobile terminal scrollback using JS scrollLines() instead of broken native scroll
+
+## 0.4.6
+
+### Patch Changes
+
+- Fix mobile keyboard scroll and layout issues:
+  - Prevent iOS Safari from scrolling the page when typing with the keyboard open (position:fixed on .app + window.scroll reset)
+  - Eliminate dead space between terminal and keyboard accessory bar by removing redundant CSS padding, tightening JS padding constant, and adding row quantization gap compensation
+  - Fix toolbar overlapping terminal content when keyboard is hidden by adding proper padding-bottom to .main, including iOS Safari bottom bar offset
+  - Strip Ink spinner bloat from terminal buffer before tailing
+  - Fix resolveCasePath priority order and suppress JSON parse warnings
+
+## 0.4.5
+
+### Patch Changes
+
+- Fix mobile keyboard toolbar positioning on iOS Safari: toolbar (Run/Stop/Run Shell) was hidden behind the accessory bar when virtual keyboard was active due to overlapping CSS positions. Remove the aggressive safety check in `updateLayoutForKeyboard()` that incorrectly dismissed keyboard state when iOS scrolled the visual viewport during typing. Add Safari-bar CSS offset to accessory bar so it properly stacks above the toolbar. Remove the double-counted Safari-bar offset when keyboard is visible since the JS transform already covers the full distance.
+
+## 0.4.4
+
+### Patch Changes
+
+- fix: mobile keyboard hides terminal content on iPhone
+
+  Fixed a bug where opening the virtual keyboard on iPhone left zero visible terminal space. Two independent mechanisms were both accounting for the keyboard height: `MobileDetection.updateAppHeight()` shrunk `--app-height` to the visual viewport height, while `KeyboardHandler.updateLayoutForKeyboard()` added a large `paddingBottom`. These double-counted, leaving negative space for the terminal (user saw accessory bar + toolbar but no terminal content).
+
+  Fix: `updateAppHeight()` now skips when the keyboard is visible, and `handleViewportResize()` restores `--app-height` to the pre-keyboard value on first detection (since MobileDetection's listener fires before KeyboardHandler's). On keyboard close, `--app-height` is re-synced to the current visual viewport.
+
+## 0.4.3
+
+### Patch Changes
+
+- Refactor case routes: extract readLinkedCases() and resolveCasePath() helpers to eliminate 6x duplicated linked-cases.json path construction and 5x duplicated file read/parse logic. Replace O(n) .some() duplicate check with O(1) Set.has() in case listing. Un-export unused isError() type guard. Standardize reply.status() to reply.code() in system routes. Update CLAUDE.md frontend module listing and SSE event count.
+
+## 0.4.2
+
+### Patch Changes
+
+- Extract monolithic app.js (~12.5K lines) into 6 focused domain modules that extend CodemanApp.prototype via Object.assign: terminal-ui.js (terminal setup, rendering pipeline, controls), respawn-ui.js (respawn banner, countdown, presets, run summary), ralph-panel.js (Ralph state panel, fix_plan, plan versioning), settings-ui.js (app settings, visibility, web push, tunnel/QR, help), panels-ui.js (subagent panel, teams, insights, file browser, log viewer), session-ui.js (quick start, session options, case settings). Fix critical deferred script init ordering bug: wrap CodemanApp instantiation in DOMContentLoaded so all defer'd mixin modules execute their Object.assign before the constructor runs. Guard missing cleanupWizardDragging() call in subagent-windows.js. Update build.mjs to minify/hash all new modules.
+
+## 0.4.1
+
+### Patch Changes
+
+- Performance optimizations: V8 compile cache for 10-20% faster cold starts, lazy-load WebGL addon (244KB saved on mobile), preload hints for critical scripts, batch tmux reconciliation (N subprocess calls → 1). Also: WebSocket session lifecycle fixes, CJK IME input support, CI upgrade to Node 24/actions v6, install.sh fork support, and CLAUDE.md/README documentation refresh.
+
+## 0.4.0
+
+### Minor Changes
+
+- Add CJK IME input textarea for xterm.js terminal (env toggle INPUT_CJK_FORM=ON). Always-visible textarea below terminal handles native browser IME composition, forwarding completed text to PTY on Enter. Supports arrow keys, Ctrl combos, backspace passthrough, and Escape to clear.
+
+  Add fork installation support to install.sh with CODEMAN_REPO_URL and CODEMAN_BRANCH env vars, allowing custom repository and branch for git clone/update operations. README updated with fork installation instructions.
+
+  Fix WebSocket session lifecycle: close WS connections when session exits (prevents orphaned listeners and stale writes to dead PTY), add readyState guard in onTerminal to stop buffering after socket closes, simplify heartbeat by removing redundant alive flag.
+
+  Add WebSocket reconnection with exponential backoff (1s-10s) on unexpected close, skipping server rejection codes (4004/4008/4009). Falls back gracefully to SSE+POST during reconnection.
+
+  Clear CJK textarea on session switch to prevent sending stale text to wrong session.
+
+## 0.3.12
+
+### Patch Changes
+
+- Add WebSocket terminal I/O with server-side DEC 2026 synchronized update markers. Replaces per-keystroke HTTP POST + SSE terminal output with a single bidirectional WebSocket connection for dramatically lower input latency. Server-side 8ms micro-batching with 16KB flush threshold groups rapid PTY events into single WS frames wrapped in DEC 2026 markers for flicker-free atomic rendering. Includes 30s ping/pong heartbeat with 10s timeout for stale connection detection through tunnels. Existing SSE + HTTP POST paths remain fully functional as transparent fallback. Resize messages validated to match HTTP route bounds (cols 1-500, rows 1-200, integers only). 16 automated route tests added for WS endpoint. Also patches 5 dependency vulnerabilities (basic-ftp, fastify, minimatch, serialize-javascript).
+
+## 0.3.11
+
+### Patch Changes
+
+- ### Session Resume & History
+  - Add `resumeSessionId` support for conversation resume after reboot
+  - Add history session resume UI and API with route shell sessions routing fix
+  - Improve session resume reliability and persist user settings across refresh
+  - Correct `claudeSessionId` for resumed sessions
+
+  ### Terminal & Frontend
+  - Upgrade xterm.js 5.3 → 6.0 with native DEC 2026 synchronized output
+  - Increase terminal scrollback from 5,000 to 20,000 lines
+  - Reduce default font size and persist tab state across refresh
+  - Resolve terminal resize scrollback ghost renders
+  - Hide subagent monitor panel by default
+
+  ### Installer
+  - Auto-detect existing install and run update instead of fresh install
+  - Auto-restart codeman-web service after update if running
+  - Show restart command when codeman-web is not a systemd service
+  - Fix one-liner restart command for background processes
+
+  ### Codebase Quality
+  - Remove dead code, consolidate imports, extract constants
+  - Repair 15 pre-existing subagent-watcher test failures
+  - Clean up DEC sync dead code
+
+## 0.3.10
+
+### Patch Changes
+
+- - feat: upgrade xterm.js from 5.3 to 6.0 with native DEC 2026 synchronized output support
+  - feat: add history session resume UI and API — resume Claude conversations after reboot
+  - feat: add resumeSessionId support for conversation resume across session restarts
+  - feat: persist active tabs across page refresh
+  - feat: improve session resume reliability and persist user settings
+  - perf: increase terminal scrollback from 5,000 to 20,000 lines
+  - fix: resolve terminal resize scrollback ghost renders
+  - fix: route shell sessions to correct endpoint on tab click
+  - fix: correct claudeSessionId for resumed sessions (use original Claude conversation ID)
+  - fix: increase default desktop font size from 12 to 14
+  - refactor: extract shared \_fetchHistorySessions() method to eliminate duplication
+  - refactor: remove dead DEC 2026 sync code (extractSyncSegments, DEC_SYNC_START/END constants)
+
+## 0.3.9
+
+### Patch Changes
+
+- Add content-hash cache busting for static assets — build step now renames JS/CSS files with MD5 content hashes (e.g. app.js → app.94b71235.js) and rewrites index.html references. HTML served with Cache-Control: no-cache so browsers always revalidate and pick up new hashed filenames after deploys. Hashed assets keep immutable 1-year cache. Eliminates the need for manual hard refresh (Ctrl+Shift+R) after deployments.
+
+  Refactor path traversal validation into shared validatePathWithinBase() helper in route-helpers.ts, replacing 6 duplicate inline checks across case-routes, plan-routes, and session-routes.
+
+  Deduplicate stripAnsi in bash-tool-parser.ts — use shared utility from utils/index.ts instead of private method.
+
+## 0.3.8
+
+### Patch Changes
+
+- Add tunnel status indicator with control panel — green pulsing dot in header when Cloudflare tunnel is active, dropdown with URL, remote clients, auth sessions, and start/stop/QR/revoke controls
+
+## 0.3.7
+
+### Patch Changes
+
+- Operation Lightspeed: 5 parallel performance optimizations — multi-layer backpressure to prevent terminal write freezes, TERMINAL_TAIL_SIZE constant with client-drop recovery, tab switching SSE gating, and local echo improvements
+- Codebase cleanup: remove dead code (unused token validation exports, PlanPhase alias), add execPattern() regex helper to eliminate repetitive .lastIndex resets, centralize 11 magic number constants into config files, fix CLAUDE.md inaccuracies, and add 316 new tests for utilities, respawn helpers, and system-routes
+
+## 0.3.6
+
+### Patch Changes
+
+- Re-enable WebGL renderer with 48KB/frame flush cap protection against GPU stalls
+
+## 0.3.5
+
+### Patch Changes
+
+- Fix Chrome "page unresponsive" crashes caused by xterm.js WebGL renderer GPU stalls during heavy terminal output. Disable WebGL by default (canvas renderer used instead), gate SSE terminal writes during tab switches, and add crash diagnostics with server-side breadcrumb collection.
+
+## 0.3.4
+
+### Patch Changes
+
+- Fix Chrome tab freeze from flicker filter buffer accumulation during active sessions, and fix shell mode feedback delay by excluding shell sessions from cursor-up filter
+
+## 0.3.3
+
+### Patch Changes
+
+- fix: eliminate WebGL re-render flicker during tab switch by keeping renderer active instead of toggling it off/on around large buffer writes
+
+## 0.3.2
+
+### Patch Changes
+
+- Make file browser panel draggable by its header
+
+## 0.3.1
+
+### Patch Changes
+
+- LLM context optimization and performance improvements: compress CLAUDE.md 21%, MEMORY.md 61%; SSE broadcast early return, cached tunnel state, cache invalidation fix, ralph todo cleanup timer; frontend SSE listener leak fix, short ID caching, subagent window handle cleanup; 100% @fileoverview coverage
+
+## 0.3.0
+
+### Minor Changes
+
+- QR code authentication for tunnel access, 7-phase codebase refactor (route extraction, type domain modules, frontend module split, config consolidation, managed timers, test infrastructure), overlay rendering fixes, and security hardening
+
 ## 0.2.9
 
 ### Patch Changes

CLAUDE.md

@@ -6,7 +6,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 | Task | Command |
 |------|---------|
-| Dev server | `npx tsx src/index.ts web` |
+| Dev server | `npm run dev` (or `npx tsx src/index.ts web`) |
 | Type check | `tsc --noEmit` |
 | Lint | `npm run lint` (fix: `npm run lint:fix`) |
 | Format | `npm run format` (check: `npm run format:check`) |
@@ -29,7 +29,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 2. **Frontend changes**: Use Playwright to load the page and assert the UI renders correctly. Use `waitUntil: 'domcontentloaded'` (not `networkidle` — SSE keeps the connection open). Wait 3-4s for polling/async data to populate, then check element visibility, text content, and CSS values
 3. **Only after verification passes**, proceed with COM
 
-The production server caches static files for 1 hour (`maxAge: '1h'` in `server.ts`). After deploying frontend changes, users may need a hard refresh (Ctrl+Shift+R) to see updates.
+The production server caches static files for 1 year (`maxAge: '1y'` in `server.ts`). After deploying frontend changes, users may need a hard refresh (Ctrl+Shift+R) to see updates.
 
 ## COM Shorthand (Deployment)
 
@@ -44,7 +44,7 @@ When user says "COM":
    "aicodeman": patch
    ---
 
-   Description of changes
+   Detailed description of ALL changes since last release (not just the most recent commit — review full git log since last version tag)
    CHANGESET
    ```
    Replace `patch` with `minor` or `major` as needed. Include `"xterm-zerolag-input": patch` on a separate line if that package changed too.
@@ -52,7 +52,7 @@ When user says "COM":
 4. **Sync CLAUDE.md version**: Update the `**Version**` line below to match the new version from `package.json`
 5. **Commit and deploy**: `git add -A && git commit -m "chore: version packages" && git push && npm run build && systemctl --user restart codeman-web`
 
-**Version**: 0.2.9 (must match `package.json`)
+**Version**: 0.5.11 (must match `package.json`)
 
 ## Project Overview
 
@@ -60,135 +60,66 @@ Codeman is a Claude Code session manager with web interface and autonomous Ralph
 
 **Tech Stack**: TypeScript (ES2022/NodeNext, strict mode), Node.js, Fastify, node-pty, xterm.js. Supports both Claude Code and OpenCode AI CLIs via pluggable CLI resolvers.
 
-**TypeScript Strictness** (see `tsconfig.json`): `noUnusedLocals`, `noUnusedParameters`, `noImplicitReturns`, `noImplicitOverride`, `noFallthroughCasesInSwitch`, `allowUnreachableCode: false`, `allowUnusedLabels: false`. Note: `src/tui` is excluded from compilation (legacy/deprecated code path).
+**TypeScript Strictness** (see `tsconfig.json`): `noUnusedLocals`, `noUnusedParameters`, `noImplicitReturns`, `noImplicitOverride`, `noFallthroughCasesInSwitch`, `allowUnreachableCode: false`, `allowUnusedLabels: false`.
 
 **Requirements**: Node.js 18+, Claude CLI, tmux
 
-## Commands
+**Git**: Main branch is `master`. SSH session chooser: `sc` (interactive), `sc 2` (quick attach), `sc -l` (list).
 
-**Note**: `npm run dev` starts the web server (equivalent to `npx tsx src/index.ts web`).
+## Additional Commands
 
-**Default port**: `3000` (web UI at `http://localhost:3000`)
+`npm run dev` = dev server. Default port: `3000`. Commands not in Quick Reference:
 
-```bash
-# Setup
-npm install                        # Install dependencies
+| Task | Command |
+|------|---------|
+| Dev with TLS | `npx tsx src/index.ts web --https` |
+| Continuous typecheck | `tsc --noEmit --watch` |
+| Test coverage | `npm run test:coverage` |
+| Production start | `npm run start` |
+| Production logs | `journalctl --user -u codeman-web -f` |
 
-# Development
-npx tsx src/index.ts web           # Dev server (RECOMMENDED)
-npx tsx src/index.ts web --https   # With TLS (only needed for remote access)
-npm run typecheck                  # Type check
-tsc --noEmit --watch               # Continuous type checking
-npm run lint                       # ESLint
-npm run lint:fix                   # ESLint with auto-fix
-npm run format                     # Prettier format
-npm run format:check               # Prettier check only
+**CI**: `.github/workflows/ci.yml` runs `typecheck`, `lint`, `format:check` on push to master/main and on PRs (Node 22). Tests excluded (they spawn tmux).
 
-# Testing (see "Testing" section for CRITICAL safety warnings)
-npx vitest run test/<file>.test.ts # Single file (SAFE)
-npx vitest run -t "pattern"        # Tests matching name
-npm run test:coverage              # With coverage report
-
-# Production
-npm run build                      # esbuild via scripts/build.mjs (not tsc)
-npm run start                      # node dist/index.js (production)
-systemctl --user restart codeman-web
-journalctl --user -u codeman-web -f
-```
-
-**CI**: `.github/workflows/ci.yml` runs `typecheck`, `lint`, and `format:check` on push to master. Tests are intentionally excluded from CI (they spawn tmux).
+**Code style**: Prettier (`singleQuote: true`, `printWidth: 120`, `trailingComma: "es5"`). ESLint flat config (`config/eslint.config.js`) allows `no-console`, warns on `@typescript-eslint/no-explicit-any`. Ignores: `app.js`, `scripts/**/*.mjs`, `src/web/public/vendor/**`, `scripts/remotion/**`.
 
 ## Common Gotchas
 
-- **Single-line prompts only** — `writeViaMux()` sends text and Enter separately; multi-line breaks Ink
-- **Don't kill tmux sessions blindly** — Check `$CODEMAN_MUX` first; you might be inside one
-- **Global regex `lastIndex` sharing** — `ANSI_ESCAPE_PATTERN_FULL/SIMPLE` have `g` flag; use `createAnsiPatternFull/Simple()` factory functions for fresh instances in loops
-- **DEC 2026 sync blocks** — Never discard incomplete sync blocks (START without END); buffer up to 50ms then flush. See `app.js:extractSyncSegments()`
-- **Terminal writes during buffer load** — Live SSE writes are queued while `_isLoadingBuffer` is true to prevent interleaving with historical data
-- **Local echo prompt scanning** — Does NOT use `buffer.cursorY` (Ink moves it); scans buffer bottom-up for visible `>` prompt marker
+- **Single-line prompts only** — `writeViaMux()` sends text+Enter separately; multi-line breaks Ink
+- **ESM only** — Never `require()`, use `await import()`. `tsx` masks CJS/ESM issues in dev but production breaks
+- **Package ≠ product name** — npm: `aicodeman`, product: **Codeman**. Release renames tags accordingly
+- **Global regex `lastIndex`** — Use `createAnsiPatternFull/Simple()` factories, not shared `g`-flag patterns in loops
 
-## Import Conventions
-
-- **Utilities**: Import from `./utils` (re-exports all): `import { LRUMap, stripAnsi } from './utils'`
-- **Types**: Use type imports: `import type { SessionState } from './types'`
-- **Config**: Import from specific files: `import { MAX_TERMINAL_BUFFER_SIZE } from './config/buffer-limits'`
+**Import conventions**: Utils from `./utils`, types from `./types` (barrel), config from specific `./config/*` files.
 
 ## Architecture
 
-### Core Files
+### Core Files (by domain)
 
-| File | Purpose |
-|------|---------|
-| `src/index.ts` | CLI entry point: global error recovery, uncaught exception guard, `MAX_CONSECUTIVE_ERRORS` auto-restart |
-| `src/session.ts` | PTY wrapper: `runPrompt()`, `startInteractive()`, `startShell()` |
-| `src/mux-interface.ts` | `TerminalMultiplexer` interface + `MuxSession` type |
-| `src/mux-factory.ts` | Create tmux multiplexer instance |
-| `src/tmux-manager.ts` | tmux session management |
-| `src/session-manager.ts` | Session lifecycle, cleanup |
-| `src/state-store.ts` | State persistence to `~/.codeman/state.json` |
-| `src/respawn-controller.ts` | State machine for autonomous cycling |
-| `src/ralph-tracker.ts` | Detects `<promise>PHRASE</promise>`, todos |
-| `src/ralph-loop.ts` | Autonomous task execution loop (polls queue, assigns tasks) |
-| `src/ralph-config.ts` | Parses `.claude/ralph-loop.local.md` plugin config |
-| `src/task.ts` | Task model for prompt execution |
-| `src/task-queue.ts` | Priority queue for tasks with dependencies |
-| `src/task-tracker.ts` | Background task tracker for subagent detection |
-| `src/subagent-watcher.ts` | Monitors Claude Code's Task tool (background agents) |
-| `src/team-watcher.ts` | Polls `~/.claude/teams/` for agent team activity; matches teams to sessions via `leadSessionId` |
-| `src/run-summary.ts` | Timeline events for "what happened while away" |
-| `src/ai-checker-base.ts` | Base class for AI-powered checkers (shared by idle + plan checkers) |
-| `src/ai-idle-checker.ts` | AI-powered idle detection |
-| `src/ai-plan-checker.ts` | AI-powered plan completion checker |
-| `src/bash-tool-parser.ts` | Parses Claude's bash tool invocations from output |
-| `src/transcript-watcher.ts` | Watches Claude's transcript files for changes |
-| `src/hooks-config.ts` | Manages `.claude/settings.local.json` hook configuration |
-| `src/push-store.ts` | VAPID key auto-gen + push subscription CRUD for Web Push |
-| `src/session-lifecycle-log.ts` | Append-only JSONL audit log at `~/.codeman/session-lifecycle.jsonl` |
-| `src/image-watcher.ts` | Watches for image file creation (screenshots, etc.) |
-| `src/file-stream-manager.ts` | Manages `tail -f` processes for live log viewing |
-| `src/plan-orchestrator.ts` | Multi-agent plan generation with research and planning phases |
-| `src/prompts/index.ts` | Barrel export for all agent prompts |
-| `src/prompts/*.ts` | Agent prompts (research-agent, planner) |
-| `src/templates/claude-md.ts` | CLAUDE.md generation for new cases |
-| `src/tunnel-manager.ts` | Manages cloudflared child process for Cloudflare tunnel remote access |
-| `src/cli.ts` | Command-line interface handlers |
-| `src/web/server.ts` | Fastify REST API + SSE at `/api/events` (~280 routes) |
-| `src/web/schemas.ts` | Zod v4 validation schemas with path/env security allowlists |
-| `src/web/public/app.js` | Frontend: xterm.js, tab management, subagent windows, mobile support (~15K lines) |
-| `src/types.ts` | All TypeScript interfaces (~70 type/interface/enum defs, ~1450 lines) |
+| Domain | Key files | Notes |
+|--------|-----------|-------|
+| **Entry** | `src/index.ts`, `src/cli.ts` | |
+| **Session** | `src/session.ts` ★, `src/session-manager.ts`, `src/session-auto-ops.ts`, `src/session-cli-builder.ts`, `src/session-lifecycle-log.ts`, `src/session-task-cache.ts` | |
+| **Mux** | `src/mux-interface.ts`, `src/mux-factory.ts`, `src/tmux-manager.ts` | |
+| **Respawn** | `src/respawn-controller.ts` ★ + 4 helpers (`-adaptive-timing`, `-health`, `-metrics`, `-patterns`) | Read `docs/respawn-state-machine.md` first |
+| **Ralph** | `src/ralph-tracker.ts` ★, `src/ralph-loop.ts` + 5 helpers (`-config`, `-fix-plan-watcher`, `-plan-tracker`, `-stall-detector`, `-status-parser`) | Read `docs/ralph-wiggum-guide.md` first |
+| **Orchestrator** | `src/orchestrator-loop.ts`, `src/orchestrator-planner.ts`, `src/orchestrator-verifier.ts` | Read `docs/orchestrator-loop-architecture.md` first |
+| **Agents** | `src/subagent-watcher.ts` ★, `src/team-watcher.ts`, `src/bash-tool-parser.ts`, `src/transcript-watcher.ts` | |
+| **AI** | `src/ai-checker-base.ts`, `src/ai-idle-checker.ts`, `src/ai-plan-checker.ts` | |
+| **Tasks** | `src/task.ts`, `src/task-queue.ts`, `src/task-tracker.ts` | |
+| **State** | `src/state-store.ts`, `src/run-summary.ts`, `src/session-lifecycle-log.ts` | |
+| **Infra** | `src/hooks-config.ts`, `src/push-store.ts`, `src/tunnel-manager.ts`, `src/image-watcher.ts`, `src/file-stream-manager.ts` | |
+| **Plan** | `src/plan-orchestrator.ts`, `src/prompts/*.ts`, `src/templates/claude-md.ts` | |
+| **Web** | `src/web/server.ts`, `src/web/sse-events.ts`, `src/web/routes/*.ts` (14 route modules + barrel), `src/web/route-helpers.ts`, `src/web/ports/*.ts`, `src/web/middleware/auth.ts`, `src/web/schemas.ts` | |
+| **Frontend** | `src/web/public/app.js` (~2.8K lines, core) + 5 infra modules (`constants.js`, `mobile-handlers.js`, `voice-input.js`, `notification-manager.js`, `keyboard-accessory.js`) + 7 domain modules (`terminal-ui.js`, `respawn-ui.js`, `ralph-panel.js`, `orchestrator-panel.js`, `settings-ui.js`, `panels-ui.js`, `session-ui.js`) + 4 feature modules (`ralph-wizard.js`, `api-client.js`, `subagent-windows.js`, `input-cjk.js`) + `sw.js` | |
+| **Types** | `src/types/index.ts` → 14 domain files | See `@fileoverview` in index.ts |
 
-**Large files** (>50KB): `app.js`, `ralph-tracker.ts`, `respawn-controller.ts`, `session.ts`, `subagent-watcher.ts` — these contain complex state machines; read `docs/respawn-state-machine.md` before modifying.
+★ = Large file (>50KB). All files have `@fileoverview` JSDoc — read that before diving in.
 
-### Local Packages
+**Local package**: `packages/xterm-zerolag-input/` — local echo overlay for xterm.js; copy embedded in `app.js`.
 
-| Package | Purpose |
-|---------|---------|
-| `packages/xterm-zerolag-input/` | Instant keystroke feedback overlay for xterm.js — eliminates perceived input latency over high-RTT connections. Source of truth for `LocalEchoOverlay`; a copy is embedded in `app.js`. Build: `npm run build` (tsup). |
+**Config**: `src/config/` — 9 files. Import from specific files, not barrel.
 
-### Config Files (`src/config/`)
-
-| File | Purpose |
-|------|---------|
-| `buffer-limits.ts` | Terminal/text buffer size limits |
-| `map-limits.ts` | Global limits for Maps, sessions, watchers |
-
-### Utilities (`src/utils/`)
-
-Re-exported via `src/utils/index.ts`. Key exports:
-
-| File | Exports |
-|------|---------|
-| `cleanup-manager.ts` | `CleanupManager` — centralized disposal for timers, intervals, watchers, listeners, streams |
-| `lru-map.ts` | `LRUMap` — bounded cache with eviction |
-| `stale-expiration-map.ts` | `StaleExpirationMap` — TTL-based map with automatic cleanup |
-| `regex-patterns.ts` | `ANSI_ESCAPE_PATTERN_FULL/SIMPLE`, `createAnsiPatternFull/Simple()`, `stripAnsi`, `TOKEN_PATTERN`, `SPINNER_PATTERN` |
-| `buffer-accumulator.ts` | `BufferAccumulator` — batches rapid writes into single flushes |
-| `claude-cli-resolver.ts` | `findClaudeDir`, `getAugmentedPath` — resolves Claude CLI paths |
-| `opencode-cli-resolver.ts` | `resolveOpenCodeDir`, `isOpenCodeAvailable` — OpenCode CLI support |
-| `string-similarity.ts` | `stringSimilarity`, `fuzzyPhraseMatch`, `todoContentHash` |
-| `token-validation.ts` | `validateTokenCounts`, `validateTokensAndCost` |
-| `nice-wrapper.ts` | `wrapWithNice` — wraps commands with `nice`/`ionice` for lower priority |
-| `type-safety.ts` | `assertNever` — exhaustive switch/case guard |
+**Utilities**: `src/utils/` — re-exported via index. Key: `CleanupManager`, `LRUMap`, `StaleExpirationMap`, `BufferAccumulator`, `stripAnsi`, `Debouncer`, `KeyedDebouncer`. Also: `claude-cli-resolver`/`opencode-cli-resolver` (CLI path resolution), `string-similarity` (fuzzy matching), `regex-patterns` (ANSI/token/spinner patterns), `assertNever` (exhaustive checks), `token-validation` (auth tokens), `nice-wrapper` (process priority).
 
 ### Data Flow
 
@@ -199,356 +130,117 @@ Re-exported via `src/utils/index.ts`. Key exports:
 
 ### Key Patterns
 
-**Input to sessions**: Use `session.writeViaMux()` for programmatic input (respawn, auto-compact). Uses tmux `send-keys -l` (literal text) + `send-keys Enter`. All prompts must be single-line.
-
-**Terminal multiplexer**: `TerminalMultiplexer` interface (`src/mux-interface.ts`) abstracts the backend. `createMultiplexer()` from `src/mux-factory.ts` creates the tmux backend.
+**Input**: `session.writeViaMux()` for programmatic input — tmux `send-keys -l` (literal) + `send-keys Enter`. Single-line only.
 
 **Idle detection**: Multi-layer (completion message → AI check → output silence → token stability). See `docs/respawn-state-machine.md`.
 
-**Token tracking**: Interactive mode parses status line ("123.4k tokens"), estimates 60/40 input/output split.
+**Hook events**: Claude Code hooks trigger via `/api/hook-event`. Key events: `permission_prompt`, `elicitation_dialog`, `idle_prompt`, `stop`, `teammate_idle`, `task_completed`. See `src/hooks-config.ts`.
 
-**Hook events**: Claude Code hooks trigger notifications via `/api/hook-event`. Key events: `permission_prompt` (tool approval needed), `elicitation_dialog` (Claude asking question), `idle_prompt` (waiting for input), `stop` (response complete), `teammate_idle` (Agent Teams), `task_completed` (Agent Teams). See `src/hooks-config.ts`.
+**Agent Teams**: `TeamWatcher` polls `~/.claude/teams/`, matches to sessions via `leadSessionId`. Teammates are in-process threads appearing as subagents. Enable: `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`. See `docs/agent-teams/`.
 
-**Web Push**: Layer 5 of the notification system. Service worker (`sw.js`) receives push events and shows OS-level notifications even when the browser tab is closed. VAPID keys auto-generated on first use and persisted to `~/.codeman/push-keys.json`. Per-subscription per-event preferences stored in `~/.codeman/push-subscriptions.json`. Expired subscriptions (410/404) auto-cleaned. Requires HTTPS or localhost. iOS requires PWA installed to home screen. See `src/push-store.ts`.
+**Circuit breaker**: Prevents respawn thrashing. States: `CLOSED` → `HALF_OPEN` → `OPEN`. Reset: `/api/sessions/:id/ralph-circuit-breaker/reset`.
 
-**Agent Teams (experimental)**: `TeamWatcher` polls `~/.claude/teams/` for team configs and matches teams to sessions via `leadSessionId`. Teammates are in-process threads (not separate OS processes) and appear as standard subagents. RespawnController checks `TeamWatcher.hasActiveTeammates()` before triggering respawn. Enable via `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` env var in `settings.local.json`. See `agent-teams/` for full docs.
+**Port interfaces**: Routes declare dependencies via port interfaces (`src/web/ports/`). Routes use intersection types (e.g., `SessionPort & EventPort`).
 
-**Circuit breaker**: Prevents respawn thrashing when Claude is stuck. States: `CLOSED` (normal) → `HALF_OPEN` (testing) → `OPEN` (blocked). Tracks consecutive no-progress, same-error-repeated, and tests-failing-too-long. Reset via API at `/api/sessions/:id/ralph-circuit-breaker/reset`.
+### Frontend
 
-**Respawn cycle metrics & health scoring**: `RespawnCycleMetrics` tracks per-cycle outcomes (success, stuck_recovery, blocked, error). `RalphLoopHealthScore` computes 0-100 health with component scores (cycleSuccess, circuitBreaker, iterationProgress, aiChecker, stuckRecovery). Available via respawn status API.
-
-**Subagent-session correlation**: Session parses Task tool output via `BashToolParser` → `SubagentWatcher` discovers new agent → calls `session.findTaskDescriptionNear()` to match description for window title.
-
-### Frontend Files
-
-| File | Purpose |
-|------|---------|
-| `src/web/public/index.html` | HTML entry point with inline critical CSS and async vendor loading |
-| `src/web/public/app.js` | Core UI: xterm.js, tab management, subagent windows, mobile support (~15K lines) |
-| `src/web/public/ralph-wizard.js` | Ralph Loop wizard UI extracted from app.js (~1K lines) |
-| `src/web/public/styles.css` | Main styling (dark theme, layout, components) |
-| `src/web/public/mobile.css` | Responsive overrides for screens <1024px (loaded conditionally via `media` attribute) |
-| `src/web/public/upload.html` | Screenshot upload page served at `/upload.html` |
-| `src/web/public/sw.js` | Service worker for Web Push notifications |
-| `src/web/public/manifest.json` | Minimal PWA manifest (required for push on Android) |
-| `src/web/public/vendor/` | Self-hosted xterm.js + addons (eliminates CDN latency) |
-
-### Frontend Architecture (`app.js`)
-
-The frontend is a single ~15K-line vanilla JS file with these key systems:
-
-| System | Key Classes/Functions | Purpose |
-|--------|----------------------|---------|
-| **Terminal rendering** | `batchTerminalWrite()`, `flushPendingWrites()`, `chunkedTerminalWrite()` | 60fps batched writes with DEC 2026 sync |
-| **Local echo overlay** | `LocalEchoOverlay` class | DOM overlay for instant mobile keystroke feedback |
-| **Mobile support** | `MobileDetection`, `KeyboardHandler`, `SwipeHandler`, `KeyboardAccessoryBar` | Touch input, viewport adaptation, swipe navigation |
-| **Subagent windows** | `openSubagentWindow()`, `closeSubagentWindow()`, `updateConnectionLines()` | Floating terminal windows with parent connection lines |
-| **Notifications** | `NotificationManager` class | 5-layer: in-app drawer, tab flash, browser API, web push, audio beep |
-| **SSE connection** | `connectSSE()`, `addListener()` | EventSource with exponential backoff (1-30s), offline queue (64KB) |
-| **Settings** | `openAppSettings()`, `apply*Visibility()` | Server-backed + localStorage persistence |
-| **Focus management** | `FocusTrap` class | Modal keyboard navigation with focus restore |
+Frontend JS modules have `@fileoverview` with `@dependency`/`@loadorder` tags. Load order: `constants.js`(1) → `mobile-handlers.js`(2) → `voice-input.js`(3) → `notification-manager.js`(4) → `keyboard-accessory.js`(5) → `input-cjk.js`(5.5) → `app.js`(6) → `terminal-ui.js`(7) → `respawn-ui.js`(8) → `ralph-panel.js`(9) → `orchestrator-panel.js`(9.5) → `settings-ui.js`(10) → `panels-ui.js`(11) → `session-ui.js`(12) → `ralph-wizard.js`(13) → `api-client.js`(14) → `subagent-windows.js`(15). `input-cjk.js` handles CJK IME composition via an always-visible textarea below the terminal (`window.cjkActive` blocks xterm's onData).
 
 **Z-index layers**: subagent windows (1000), plan agents (1100), log viewers (2000), image popups (3000), local echo overlay (7).
 
-**Built-in respawn presets**: `solo-work` (3s idle, 60min), `subagent-workflow` (45s idle, 240min), `team-lead` (90s idle, 480min), `ralph-todo` (8s idle, 480min, works through @fix_plan.md tasks), `overnight-autonomous` (10s idle, 480min, full reset).
+**Respawn presets**: `solo-work` (3s/60min), `subagent-workflow` (45s/240min), `team-lead` (90s/480min), `ralph-todo` (8s/480min), `overnight-autonomous` (10s/480min).
 
-**Keyboard shortcuts**: Escape (close panels), Ctrl+? (help), Ctrl+Enter (quick start), Ctrl+W (kill session), Ctrl+Tab (next session), Ctrl+K (kill all), Ctrl+L (clear), Ctrl+Shift+R (restore size), Ctrl/Cmd +/- (font size).
+**Keyboard shortcuts**: Escape (close), Ctrl+? (help), Ctrl+W (kill), Ctrl+Tab (next), Alt+1-9 (switch tab), Shift+Enter (newline), Ctrl+L (clear), Ctrl+Shift+R (restore size), Ctrl/Cmd +/- (font).
 
 ### Security
 
-- **HTTP Basic Auth**: Optional via `CODEMAN_USERNAME`/`CODEMAN_PASSWORD` env vars
-- **Session cookies**: After Basic Auth, a 24h session cookie (`codeman_session`) is issued so credentials aren't re-sent on every request. Active sessions auto-extend. SSE works via same-origin cookie (`EventSource` can't send custom headers).
-- **Rate limiting**: 10 failed auth attempts per IP triggers 429 rejection (15-minute decay window). Manual `StaleExpirationMap` counter — no `@fastify/rate-limit` needed.
-- **Hook bypass**: `/api/hook-event` POST is exempt from auth — Claude Code hooks curl this from localhost and can't present credentials. Safe: validated by `HookEventSchema`, only triggers broadcasts.
-- **CORS**: Restricted to localhost only
-- **Security headers**: X-Content-Type-Options, X-Frame-Options, CSP; HSTS if HTTPS
-- **Path validation** (`schemas.ts`): Strict allowlist regex, no shell metacharacters, no traversal, must be absolute
-- **Env var allowlist**: Only `CLAUDE_CODE_*` prefixes allowed; blocks `PATH`, `LD_PRELOAD`, `NODE_OPTIONS`, `CODEMAN_*` keys
-- **File streaming TOCTOU protection**: `FileStreamManager` calls `realpathSync()` twice (at validation and before spawn) to catch symlink swaps
+| Layer | Details |
+|-------|---------|
+| **Auth** | Optional HTTP Basic via `CODEMAN_USERNAME`/`CODEMAN_PASSWORD` env vars |
+| **QR Auth** | Single-use 6-char tokens (60s TTL) for tunnel login. See `docs/qr-auth-plan.md` |
+| **Sessions** | 24h cookie (`codeman_session`), auto-extend, device context audit |
+| **Rate limit** | 10 failed auth/IP → 429 (15min decay). QR has separate limiter |
+| **Hook bypass** | `/api/hook-event` exempt from auth (localhost-only, schema-validated) |
+| **Env vars** | `CODEMAN_MUX` (managed session), `CODEMAN_API_URL` (auto-set for hooks) |
+| **Validation** | Zod schemas, path allowlist regex, `CLAUDE_CODE_*` env prefix allowlist |
+| **Headers** | CORS localhost-only, CSP, X-Frame-Options, HSTS if HTTPS |
 
-### SSE Event Categories
+### SSE Event Registry
 
-~80+ event types broadcast via `broadcast()`. Key categories:
+~117 event types in `src/web/sse-events.ts` (backend) and `SSE_EVENTS` in `constants.js` (frontend). Both must be kept in sync.
 
-| Category | Events | Purpose |
-|----------|--------|---------|
-| Session | `session:created/updated/deleted/working/idle/exit/error/completion` | Lifecycle |
-| Terminal | `session:terminal`, `session:clearTerminal`, `session:needsRefresh` | Output streaming |
-| Respawn | `respawn:stateChanged/cycleStarted/blocked/aiCheck*/planCheck*/timer*` | Respawn state machine |
-| Subagent | `subagent:discovered/updated/completed/tool_call/progress` | Background agents |
-| Ralph | `session:ralphLoopUpdate/ralphTodoUpdate/ralphCompletionDetected` | Ralph tracking |
-| Hooks | `hook:{eventName}` (dynamic) | Claude Code hook events |
-| Plan | `plan:started/progress/completed/cancelled/subagent` | Plan orchestration |
-| Mux | `mux:created/killed/died/statsUpdated` | tmux process monitor |
-| Image | `image:detected` | Screenshot detection |
+### API Routes
 
-### API Route Categories
-
-~280 route handlers in `server.ts:buildServer()`. Key groups:
-
-| Group | Prefix | Count | Key endpoints |
-|-------|--------|-------|---------------|
-| Sessions | `/api/sessions` | ~20 | CRUD, input, resize, interactive, shell |
-| Respawn | `/api/sessions/:id/respawn` | 5 | start, stop, enable, config |
-| Ralph | `/api/sessions/:id/ralph-*` | 6 | state, status, config, circuit-breaker |
-| Plan | `/api/sessions/:id/plan/*` | 5 | task CRUD, checkpoint, history, rollback |
-| Subagents | `/api/subagents` | 7 | list, transcript, kill, cleanup |
-| Cases | `/api/cases` | 5 | CRUD, link, fix-plan |
-| Scheduled | `/api/scheduled` | 4 | CRUD for scheduled runs |
-| Push | `/api/push` | 4 | VAPID key, subscribe, update prefs, unsubscribe |
-| System | `/api/status`, `/api/stats`, `/api/config`, `/api/settings` | 8 | App state, config |
-| Files | `/api/sessions/:id/file*`, `tail-file` | 5 | Browser, preview, raw, tail stream |
-| Mux | `/api/mux-sessions` | 4 | tmux management, stats |
+~124 handlers across 14 route files in `src/web/routes/`: system (36), sessions (25), orchestrator (10), ralph (9), plan (8), respawn (7), cases (7), files (5), mux (5), scheduled (4), push (4), teams (2), hooks (1), ws (1 WebSocket). Each file has `@fileoverview` with endpoint details.
 
 ## Adding Features
 
-- **API endpoint**: Types in `types.ts`, route in `server.ts:buildServer()`, use `createErrorResponse()`. Validate request bodies with Zod schemas in `schemas.ts`.
-- **SSE event**: Emit via `broadcast()`, handle in `app.js` SSE listener section (search `addListener(`)
-- **Session setting**: Add to `SessionState` in `types.ts`, include in `session.toState()`, call `persistSessionState()`
-- **Hook event**: Add to `HookEventType` in `types.ts`, add hook command in `hooks-config.ts:generateHooksConfig()`, update `HookEventSchema` in `schemas.ts`
-- **Mobile feature**: Add to relevant mobile singleton (`KeyboardHandler`, `KeyboardAccessoryBar`, etc.), test with `MobileDetection.isMobile()` guard
-- **New test**: Pick unique port (search `const PORT =`), add port comment to test file header. Tests use ports 3150+.
+- **API endpoint**: Types in `src/types/` domain file, route in `src/web/routes/*-routes.ts`, use `createErrorResponse()`. Validate with Zod schemas in `schemas.ts`.
+- **SSE event**: Add to `src/web/sse-events.ts` + `SSE_EVENTS` in `constants.js`, emit via `broadcast()`, handle in `app.js` (`addListener(`)
+- **Session setting**: Add to `SessionState`, include in `session.toState()`, call `persistSessionState()`
+- **Hook event**: Add to `HookEventType`, add hook in `hooks-config.ts:generateHooksConfig()`, update `HookEventSchema`
+- **Mobile feature**: Add to relevant singleton, guard with `MobileDetection.isMobile()`
+- **New test**: Pick unique port (search `const PORT =`). Route tests use `app.inject()` (no port needed) — see `test/routes/_route-test-utils.ts`.
 
-**Validation**: Uses Zod v4 for request validation. Define schemas in `schemas.ts` and use `.parse()` or `.safeParse()`. Note: Zod v4 has different API from v3 (e.g., `z.object()` options changed, error formatting differs).
+**Validation**: Zod v4 (different API from v3). Define schemas in `schemas.ts`, use `.parse()`/`.safeParse()`.
 
 ## State Files
 
-| File | Purpose |
-|------|---------|
-| `~/.codeman/state.json` | Sessions, settings, tokens, respawn config |
-| `~/.codeman/mux-sessions.json` | Tmux session metadata for recovery |
-| `~/.codeman/settings.json` | User preferences |
-| `~/.codeman/push-keys.json` | VAPID key pair for Web Push (auto-generated) |
-| `~/.codeman/push-subscriptions.json` | Registered push notification subscriptions |
-
-## Default Settings
-
-UI defaults are set in `src/web/public/app.js` using `??` fallbacks. To change defaults, edit `openAppSettings()` and `apply*Visibility()` functions.
-
-**Key defaults:** Most panels hidden (monitor, subagents shown), notifications enabled (audio disabled), subagent tracking on, Ralph tracking off.
+All in `~/.codeman/`: `state.json` (sessions, settings, respawn), `mux-sessions.json` (tmux recovery), `settings.json` (user prefs), `push-keys.json` (VAPID), `push-subscriptions.json`, `session-lifecycle.jsonl` (audit log).
 
 ## Testing
 
-**CRITICAL: You are running inside a Codeman-managed tmux session.** Never run `npx vitest run` (full suite) — it spawns/kills tmux sessions and will crash your own session. Instead:
+**CRITICAL: You are running inside a Codeman-managed tmux session.** Never run `npx vitest run` (full suite) — it spawns/kills tmux sessions and will crash your own session. Only run individual files:
 
 ```bash
-# Safe: run individual test files
-npx vitest run test/<specific-file>.test.ts
-
-# Safe: run tests matching a pattern
-npx vitest run -t "pattern"
-
-# DANGEROUS from inside Codeman — will kill your tmux session:
-# npx vitest run          ← DON'T DO THIS
+npx vitest run test/<specific-file>.test.ts     # Single file (SAFE)
+npx vitest run -t "pattern"                      # By name (SAFE)
+# npx vitest run                                 # DANGEROUS — DON'T DO THIS
 ```
 
-**Ports**: Unit tests pick unique ports manually. Search `const PORT =` before adding new tests.
+**Config**: Vitest with `globals: true`, `fileParallelism: false`. Timeout 30s, teardown 60s.
 
-**Config**: Vitest with `globals: true`, `fileParallelism: false`. Unit timeout 30s.
+**Safety**: `test/setup.ts` snapshots pre-existing tmux sessions and never kills them. Only `registerTestTmuxSession()` sessions get cleaned up.
 
-**Safety**: `test/setup.ts` snapshots pre-existing tmux sessions at load time and never kills them. Only sessions registered via `registerTestTmuxSession()` get cleaned up.
+**Ports**: Pick unique ports manually. Search `const PORT =` before adding new tests.
 
-**Respawn tests**: Use MockSession from `test/respawn-test-utils.ts` to avoid spawning real Claude processes.
+**Respawn tests**: Use `MockSession` from `test/respawn-test-utils.ts`. **Route tests**: `app.inject()` in `test/routes/`. **Mobile tests**: Playwright suite in `test/mobile/` (135 device profiles).
 
-**Mobile tests**: Separate Playwright-based suite in `mobile-test/` with 135 device profiles. Run via `npx vitest run --config mobile-test/vitest.config.ts`. See `mobile-test/README.md`.
+## Screenshots
 
-## Screenshots ("sc")
-
-When the user says "check the sc", "screenshot", or "sc", they mean uploaded screenshots from their mobile device. Screenshots are saved to `~/.codeman/screenshots/` and uploaded via `/upload.html` on the Codeman web UI. To view them, use the Read tool on the image files:
-
-```bash
-ls ~/.codeman/screenshots/        # List uploaded screenshots
-# Then use Read tool on individual files — Claude Code can view images natively
-```
-
-API: `GET /api/screenshots` (list), `GET /api/screenshots/:name` (serve), `POST /api/screenshots` (upload multipart/form-data). Source: `src/web/public/upload.html`.
+Mobile screenshots in `~/.codeman/screenshots/`. API: `GET /api/screenshots`, `POST /api/screenshots`.
 
 ## Debugging
 
 ```bash
-tmux list-sessions                  # List tmux sessions
-tmux attach-session -t <name>       # Attach (Ctrl+B D to detach)
-curl localhost:3000/api/sessions    # Check sessions
-curl localhost:3000/api/status | jq # Full app state
-cat ~/.codeman/state.json | jq    # View persisted state
-curl localhost:3000/api/subagents   # List background agents
-curl localhost:3000/api/sessions/:id/run-summary | jq  # Session timeline
+tmux list-sessions                                 # List tmux sessions
+curl localhost:3000/api/sessions | jq              # Check sessions
+curl localhost:3000/api/status | jq                # Full app state
+curl localhost:3000/api/subagents | jq             # Background agents
+cat ~/.codeman/state.json | jq                     # Persisted state
 ```
 
-## Troubleshooting
+## Performance & Limits
 
-| Problem | Check | Fix |
-|---------|-------|-----|
-| Session won't start | `tmux list-sessions` for orphans | Kill orphaned sessions, check Claude CLI installed |
-| Port 3000 in use | `lsof -i :3000` | Kill conflicting process or use `--port` flag |
-| SSE not connecting | Browser console for errors | Check CORS, ensure server running |
-| Respawn not triggering | Session settings → Respawn enabled? | Enable respawn, check idle timeout config |
-| Terminal blank on tab switch | Network tab for `/api/sessions/:id/buffer` | Check session exists, restart server |
-| Tests failing on session limits | `tmux list-sessions \| wc -l` | Clean up: `tmux list-sessions \| grep test \| awk -F: '{print $1}' \| xargs -I{} tmux kill-session -t {}` |
-| State not persisting | `cat ~/.codeman/state.json` | Check file permissions, disk space |
+Target: 20 sessions, 50 agent windows at 60fps. Limits in `src/config/`: terminal 2MB, text 1MB, messages 1000, max agents 500, max sessions 50, max SSE clients 100. Use `LRUMap` for bounded caches, `StaleExpirationMap` for TTL cleanup. Anti-flicker pipeline: `docs/terminal-anti-flicker.md`.
 
-## Performance Constraints
+## References
 
-The app must stay fast with 20 sessions and 50 agent windows:
-- 60fps terminal (16ms batching + `requestAnimationFrame`)
-- Auto-trimming buffers (2MB terminal max)
-- Debounced state persistence (500ms)
-- SSE adaptive batching: 16ms (normal), 32ms (moderate), 50ms (rapid); immediate flush at 32KB
-- SSE backpressure handling: skip writes to backpressured clients, recover via `session:needsRefresh` on drain
-- Cached endpoints: `/api/sessions` and `/api/status` use 1s TTL caches to avoid expensive serialization
-- Frontend buffer loads: 128KB chunks via `requestAnimationFrame` to prevent UI jank
-
-## Terminal Anti-Flicker System
-
-Claude Code uses Ink (React for terminals), which redraws the screen on every state change. Codeman implements a 6-layer anti-flicker pipeline for smooth 60fps output:
-
-```
-PTY Output → Server Batching (16-50ms) → DEC 2026 Wrap → SSE → Client rAF → xterm.js
-```
-
-**Key functions:** `server.ts:batchTerminalData()`, `server.ts:flushTerminalBatches()`, `app.js:batchTerminalWrite()`, `app.js:extractSyncSegments()`
-
-**Typical latency:** 16-32ms. Optional per-session flicker filter adds ~50ms for problematic terminals.
-
-See `docs/terminal-anti-flicker.md` for full implementation details (adaptive batching, DEC 2026 markers, edge cases).
-
-## Resource Limits
-
-Limits are centralized in `src/config/buffer-limits.ts` and `src/config/map-limits.ts`.
-
-**Buffer limits** (per session):
-| Buffer | Max | Trim To |
-|--------|-----|---------|
-| Terminal | 2MB | 1.5MB |
-| Text output | 1MB | 768KB |
-| Messages | 1000 | 800 |
-
-**Map limits** (global):
-| Resource | Max |
-|----------|-----|
-| Tracked agents | 500 |
-| Concurrent sessions | 50 |
-| SSE clients total | 100 |
-| File watchers | 500 |
-
-Use `LRUMap` for bounded caches with eviction, `StaleExpirationMap` for TTL-based cleanup.
-
-## Where to Find More Information
-
-| Topic | Location |
-|-------|----------|
-| **Respawn state machine** | `docs/respawn-state-machine.md` |
-| **Ralph Loop guide** | `docs/ralph-wiggum-guide.md` |
-| **Claude Code hooks** | `docs/claude-code-hooks-reference.md` |
-| **Terminal anti-flicker** | `docs/terminal-anti-flicker.md` |
-| **Agent Teams (experimental)** | `agent-teams/README.md`, `agent-teams/design.md` |
-| **API routes** | `src/web/server.ts:buildServer()` or README.md |
-| **SSE events** | Search `broadcast(` in `server.ts` |
-| **Session statuses** | `SessionStatus` type in `src/types.ts` |
-| **Error codes** | `createErrorResponse()` in `src/types.ts` |
-| **Test utilities** | `test/respawn-test-utils.ts` |
-| **Mobile test suite** | `mobile-test/README.md` |
-| **OpenCode integration** | `docs/opencode-integration.md` |
-| **Local echo overlay** | `docs/local-echo-overlay-plan.md` |
-| **Performance investigation** | `docs/performance-investigation-report.md` |
-| **First-load optimization** | `docs/first-load-optimization-plan.md`, `docs/perf-audit-first-load.md` |
-| **Dead code audit** | `docs/cleanup-findings.md` |
-| **TypeScript improvements** | `docs/typescript-improvement-suggestions.md` |
-| **Browser testing** | `docs/browser-testing-guide.md` |
-| **Mobile testing report** | `docs/mobile-testing-report.md` |
-| **Voice input** | `docs/voice-input-plan.md` |
-| **Improvement roadmaps** | `docs/respawn-improvement-plan.md`, `docs/ralph-improvement-plan.md`, `docs/plan-improvement-roadmap.md` |
-| **Background keystroke forwarding** | `docs/background-keystroke-forwarding-merged-plan.md` |
-| **Run summary** | `docs/run-summary-plan.md` |
-
-Additional design docs and investigation reports are in the `docs/` directory.
+Deep-dive docs in `docs/`: `respawn-state-machine.md`, `ralph-wiggum-guide.md`, `claude-code-hooks-reference.md`, `terminal-anti-flicker.md`, `opencode-integration.md`, `qr-auth-plan.md`, `orchestrator-loop-architecture.md`, `browser-testing-guide.md`. Agent Teams: `docs/agent-teams/README.md`. SSE events: `src/web/sse-events.ts` + `constants.js`.
 
 ## Scripts
 
-| Script | Purpose |
-|--------|---------|
-| `scripts/tmux-manager.sh` | Safe tmux session management (use instead of direct kill commands) |
-| `scripts/monitor-respawn.sh` | Monitor respawn state machine in real-time |
-| `scripts/watch-subagents.ts` | Real-time subagent transcript watcher (list, follow by session/agent ID) |
-| `scripts/codeman-web.service` | systemd service file for production deployment |
-| `scripts/codeman-tunnel.service` | systemd service file for persistent Cloudflare tunnel |
-| `scripts/tunnel.sh` | Start/stop/check Cloudflare quick tunnel (`./scripts/tunnel.sh start\|stop\|url`) |
-| `scripts/build.mjs` | esbuild-based production build (called by `npm run build`) |
-| `scripts/postinstall.js` | npm postinstall hook for setup |
-
-Additional scripts in `scripts/` for screenshots, demos, Ralph wizards, and browser testing.
+Key: `scripts/tmux-manager.sh` (safe tmux mgmt), `scripts/tunnel.sh` (tunnel start/stop/url). Production: `scripts/codeman-web.service`, `scripts/codeman-tunnel.service`.
 
 ## Memory Leak Prevention
 
-Frontend runs long (24+ hour sessions); all Maps/timers must be cleaned up.
-
-### Cleanup Patterns
-When adding new event listeners or timers:
-1. Store handler references for later removal
-2. Add cleanup to appropriate `stop()` or `cleanup*()` method
-3. For singleton watchers, store refs in class properties and remove in server `stop()`
-
-**Backend**: Clear Maps in `stop()`, null promise callbacks on error, remove watcher listeners on shutdown. Use `CleanupManager` for centralized disposal — supports timers, intervals, watchers, listeners, streams. Guard async callbacks with `if (this.cleanup.isStopped) return`.
-
-**Frontend**: Store drag/resize handlers on elements, clean up in `close*()` functions. SSE reconnect calls `handleInit()` which resets state. SSE listeners are tracked in an array and removed on reconnect to prevent accumulation.
-
-Run `npx vitest run test/memory-leak-prevention.test.ts` to verify patterns.
+24+ hour sessions: use `CleanupManager`, clear Maps in `stop()`, guard async with `if (this.cleanup.isStopped) return`. Frontend: store handler refs, clean in `close*()`. Verify: `npx vitest run test/memory-leak-prevention.test.ts`.
 
 ## Common Workflows
 
-**Investigating a bug**: Start dev server (`npx tsx src/index.ts web`), reproduce in browser, check terminal output and `~/.codeman/state.json` for clues.
+**Bug investigation**: Dev server → reproduce in browser → check terminal + `~/.codeman/state.json`.
+**Respawn changes**: Read `docs/respawn-state-machine.md` first. Use `MockSession` from `test/respawn-test-utils.ts`.
 
-**Adding a new API endpoint**: Define types in `types.ts`, add route in `server.ts:buildServer()`, broadcast SSE events if needed, handle in `app.js:handleSSEEvent()`.
+## Tunnel
 
-**Modifying respawn behavior**: Study `docs/respawn-state-machine.md` first. The state machine is in `respawn-controller.ts`. Use MockSession from `test/respawn-test-utils.ts` for testing.
-
-**Modifying mobile behavior**: Mobile singletons (`MobileDetection`, `KeyboardHandler`, `SwipeHandler`, `KeyboardAccessoryBar`) all have `init()`/`cleanup()` lifecycle. KeyboardHandler uses `visualViewport` API for iOS keyboard detection (100px threshold for address bar drift). All mobile handlers are re-initialized after SSE reconnect to prevent stale closures.
-
-**Adding a file watcher**: Use `ImageWatcher` as a template pattern — chokidar with `awaitWriteFinish`, burst throttling (max 20/10s), debouncing (200ms), and auto-ignore of `node_modules/.git/dist/`.
-
-## Tunnel Setup (Remote Access)
-
-Access Codeman from mobile/remote devices via Cloudflare quick tunnel.
-
-```
-Browser → Cloudflare Edge (HTTPS) → cloudflared → localhost:3000
-```
-
-**Prerequisites**: `cloudflared` installed (`cloudflared --version`), `CODEMAN_PASSWORD` set in environment.
-
-### Quick Start
-
-```bash
-# Via CLI
-./scripts/tunnel.sh start      # Start tunnel, prints public URL
-./scripts/tunnel.sh url        # Show current URL
-./scripts/tunnel.sh stop       # Stop tunnel
-
-# Via web UI: Settings → Tunnel → Toggle On
-```
-
-### systemd Service (Persistent)
-
-```bash
-# Install and enable
-cp scripts/codeman-tunnel.service ~/.config/systemd/user/
-systemctl --user daemon-reload
-systemctl --user enable --now codeman-tunnel
-
-# Check logs
-journalctl --user -u codeman-tunnel -f
-```
-
-### Auth Flow
-
-1. First request → browser shows Basic Auth prompt (username: `admin` or `CODEMAN_USERNAME`)
-2. On success → server issues `codeman_session` HttpOnly cookie (24h TTL, auto-extends on activity)
-3. Subsequent requests → cookie authenticates silently (no more prompts)
-4. SSE works automatically — `EventSource` sends same-origin cookies
-5. 10 failed attempts per IP → 429 rate limit (15-minute decay)
-
-### Security Requirements
-
-- **Always set `CODEMAN_PASSWORD`** before exposing via tunnel — without it, anyone with the URL has full access
-- Session cookies are `Secure` when using `--https` flag; through Cloudflare tunnel without `--https`, cookies are non-Secure but traffic is still encrypted end-to-end via Cloudflare
-- `/api/hook-event` bypasses auth (localhost-only Claude Code hooks need unauthenticated access)
+`./scripts/tunnel.sh start|stop|url`. **Always set `CODEMAN_PASSWORD`** before exposing via tunnel.

README.md

@@ -11,7 +11,7 @@
 <p align="center">
   <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-1e3a5f?style=flat-square" alt="License: MIT"></a>
   <a href="https://nodejs.org/"><img src="https://img.shields.io/badge/Node.js-18%2B-22c55e?style=flat-square&logo=node.js&logoColor=white" alt="Node.js 18+"></a>
-  <a href="https://www.typescriptlang.org/"><img src="https://img.shields.io/badge/TypeScript-5.5-3b82f6?style=flat-square&logo=typescript&logoColor=white" alt="TypeScript 5.5"></a>
+  <a href="https://www.typescriptlang.org/"><img src="https://img.shields.io/badge/TypeScript-5.9-3b82f6?style=flat-square&logo=typescript&logoColor=white" alt="TypeScript 5.9"></a>
   <a href="https://fastify.dev/"><img src="https://img.shields.io/badge/Fastify-5.x-1e3a5f?style=flat-square&logo=fastify&logoColor=white" alt="Fastify"></a>
   <img src="https://img.shields.io/badge/Tests-1435%20total-22c55e?style=flat-square" alt="Tests">
 </p>
@@ -28,29 +28,67 @@
 curl -fsSL https://raw.githubusercontent.com/Ark0N/Codeman/master/install.sh | bash
 ```
 
-This installs Node.js and tmux if missing, clones Codeman to `~/.codeman/app`, and builds it. You'll need at least one AI coding CLI installed — [Claude Code](https://docs.anthropic.com/en/docs/claude-code) or [OpenCode](https://opencode.ai) (or both). After install:
+This installs Node.js and tmux if missing, clones Codeman to `~/.codeman/app`, and builds it.
+
+You'll need at least one AI coding CLI installed — [Claude Code](https://docs.anthropic.com/en/docs/claude-code) or [OpenCode](https://opencode.ai) (or both). After install:
 
 ```bash
 codeman web
 # Open http://localhost:3000 — press Ctrl+Enter to start your first session
 ```
 
-**Update to latest version:**
-```bash
-curl -fsSL https://raw.githubusercontent.com/Ark0N/Codeman/master/install.sh | bash -s update
-```
-
 <details>
 <summary><strong>Run as a background service</strong></summary>
 
 **Linux (systemd):**
 ```bash
-mkdir -p ~/.config/systemd/user && printf '[Unit]\nDescription=Codeman Web Server\nAfter=network.target\n\n[Service]\nType=simple\nExecStart=%s %s/dist/index.js web\nRestart=always\nRestartSec=10\n\n[Install]\nWantedBy=default.target\n' "$(which node)" "$HOME/.codeman/app" > ~/.config/systemd/user/codeman-web.service && systemctl --user daemon-reload && systemctl --user enable --now codeman-web && loginctl enable-linger $USER
+mkdir -p ~/.config/systemd/user
+cat > ~/.config/systemd/user/codeman-web.service << EOF
+[Unit]
+Description=Codeman Web Server
+After=network.target
+
+[Service]
+Type=simple
+ExecStart=$(which node) $HOME/.codeman/app/dist/index.js web
+Restart=always
+RestartSec=10
+
+[Install]
+WantedBy=default.target
+EOF
+systemctl --user daemon-reload
+systemctl --user enable --now codeman-web
+loginctl enable-linger $USER
 ```
 
 **macOS (launchd):**
 ```bash
-mkdir -p ~/Library/LaunchAgents && printf '<?xml version="1.0" encoding="UTF-8"?>\n<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">\n<plist version="1.0"><dict><key>Label</key><string>com.codeman.web</string><key>ProgramArguments</key><array><string>%s</string><string>%s/dist/index.js</string><string>web</string></array><key>RunAtLoad</key><true/><key>KeepAlive</key><true/><key>StandardOutPath</key><string>/tmp/codeman.log</string><key>StandardErrorPath</key><string>/tmp/codeman.log</string></dict></plist>\n' "$(which node)" "$HOME/.codeman/app" > ~/Library/LaunchAgents/com.codeman.web.plist && launchctl load ~/Library/LaunchAgents/com.codeman.web.plist
+mkdir -p ~/Library/LaunchAgents
+cat > ~/Library/LaunchAgents/com.codeman.web.plist << EOF
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
+  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+  <key>Label</key>
+  <string>com.codeman.web</string>
+  <key>ProgramArguments</key>
+  <array>
+    <string>$(which node)</string>
+    <string>$HOME/.codeman/app/dist/index.js</string>
+    <string>web</string>
+  </array>
+  <key>RunAtLoad</key><true/>
+  <key>KeepAlive</key><true/>
+  <key>StandardOutPath</key>
+  <string>/tmp/codeman.log</string>
+  <key>StandardErrorPath</key>
+  <string>/tmp/codeman.log</string>
+</dict>
+</plist>
+EOF
+launchctl bootstrap gui/$(id -u) ~/Library/LaunchAgents/com.codeman.web.plist
 ```
 </details>
 
@@ -68,11 +106,23 @@ Codeman requires tmux, so Windows users need [WSL](https://learn.microsoft.com/e
 
 ## Mobile-Optimized Web UI
 
-The most responsive AI coding agent experience on any phone. Full xterm.js terminal with local echo, swipe navigation, and a touch-optimized interface designed for real remote work.
+The most responsive AI coding agent experience on any phone. Full xterm.js terminal with local echo, swipe navigation, and a touch-optimized interface designed for real remote work — not a desktop UI crammed onto a small screen.
+
+<table>
+<tr>
+<td align="center" width="33%"><img src="docs/screenshots/mobile-landing-qr.png" alt="Mobile — landing page with QR auth" width="260"></td>
+<td align="center" width="33%"><img src="docs/screenshots/mobile-session-idle.png" alt="Mobile — idle session with keyboard accessory" width="260"></td>
+<td align="center" width="33%"><img src="docs/screenshots/mobile-session-active.png" alt="Mobile — active agent session" width="260"></td>
+</tr>
+<tr>
+<td align="center"><em>Landing page with QR auth</em></td>
+<td align="center"><em>Keyboard accessory bar</em></td>
+<td align="center"><em>Agent working in real-time</em></td>
+</tr>
+</table>
 
 <table>
 <tr>
-<td rowspan="8" width="320"><img src="docs/screenshots/mobile-keyboard-open.png" alt="Mobile — keyboard open" width="300"></td>
 <th>Terminal Apps</th>
 <th>Codeman Mobile</th>
 </tr>
@@ -82,11 +132,22 @@ The most responsive AI coding agent experience on any phone. Full xterm.js termi
 <tr><td>No notifications</td><td>Push alerts for approvals and idle</td></tr>
 <tr><td>Manual reconnect</td><td>tmux persistence</td></tr>
 <tr><td>No agent visibility</td><td>Background agents in real-time</td></tr>
-<tr><td>Copy-paste slash commands</td><td>One-tap <code>/init</code></tr>
+<tr><td>Copy-paste slash commands</td><td>One-tap <code>/init</code>, <code>/clear</code>, <code>/compact</code></td></tr>
+<tr><td>Password typing on phone</td><td><b>QR code scan — instant auth</b></td></tr>
 </table>
 
-- **Swipe navigation** — left/right on the terminal to switch sessions (80px threshold, 300ms)
+### Secure QR Code Authentication
+
+Typing passwords on a phone keyboard is miserable. Codeman replaces it with **cryptographically secure single-use QR tokens** — scan the code displayed on your desktop and your phone is authenticated instantly.
+
+Each QR encodes a URL containing a 6-character short code that maps to a 256-bit secret (`crypto.randomBytes(32)`) on the server. Tokens auto-rotate every **60 seconds**, are **atomically consumed on first scan** (replays always fail), and use **hash-based `Map.get()` lookup** that leaks nothing through response timing. The short code is an opaque pointer — the real secret never appears in browser history, `Referer` headers, or Cloudflare edge logs.
+
+The security design addresses all 6 critical QR auth flaws identified in ["Demystifying the (In)Security of QR Code-based Login"](https://www.usenix.org/conference/usenixsecurity25/presentation/zhang-xin) (USENIX Security 2025, which found 47 of the top-100 websites vulnerable): single-use enforcement, short TTL, cryptographic randomness, server-side generation, real-time desktop notification on scan (QRLjacking detection), and IP + User-Agent session binding with manual revocation. Dual-layer rate limiting (per-IP + global) makes brute force infeasible across 62^6 = 56.8 billion possible codes. Full security analysis: [`docs/qr-auth-plan.md`](docs/qr-auth-plan.md)
+
+### Touch-Optimized Interface
+
 - **Keyboard accessory bar** — `/init`, `/clear`, `/compact` quick-action buttons above the virtual keyboard. Destructive commands (`/clear`, `/compact`) require a double-press to confirm — first tap arms the button, second tap executes — so you never fire one by accident on a bumpy commute
+- **Swipe navigation** — left/right on the terminal to switch sessions (80px threshold, 300ms)
 - **Smart keyboard handling** — toolbar and terminal shift up when keyboard opens (uses `visualViewport` API with 100px threshold for iOS address bar drift)
 - **Safe area support** — respects iPhone notch and home indicator via `env(safe-area-inset-*)`
 - **44px touch targets** — all buttons meet iOS Human Interface Guidelines minimum sizes
@@ -120,6 +181,10 @@ Watch background agents work in real-time. Codeman monitors agent activity and d
 
 ## Zero-Lag Input Overlay
 
+<p align="center">
+  <img src="docs/images/zerolag-demo.gif" alt="Zerolag Demo — local echo vs server echo side-by-side" width="900">
+</p>
+
 When accessing your coding agent remotely (VPN, Tailscale, SSH tunnel), every keystroke normally takes 200-300ms to round-trip. Codeman implements a **Mosh-inspired local echo system** that makes typing feel instant regardless of latency.
 
 A pixel-perfect DOM overlay inside xterm.js renders keystrokes at 0ms. Background forwarding silently sends every character to the PTY in 50ms debounced batches, so Tab completion, `Ctrl+R` history search, and all shell features work normally. When the server echo arrives 200-300ms later, the overlay seamlessly disappears and the real terminal text takes over — the transition is invisible.
@@ -239,6 +304,80 @@ loginctl enable-linger $USER
 
 </details>
 
+### QR Code Authentication
+
+Typing a password on a phone keyboard is terrible. Codeman solves this with **ephemeral single-use QR tokens** — scan the code on your desktop, and your phone is instantly authenticated. No password prompt, no typing, no clipboard.
+
+```
+Desktop displays QR  →  Phone scans  →  GET /q/Xk9mQ3  →  Server validates
+→  Token atomically consumed (single-use)  →  Session cookie issued  →  302 to /
+→  Desktop notified: "Device authenticated via QR"  →  New QR auto-generated
+```
+
+Someone who only has the bare tunnel URL (without the QR) still hits the standard password prompt. The QR is the fast path; the password is the fallback.
+
+#### How It Works
+
+The server maintains a rotating pool of short-lived, single-use tokens. Each token consists of a 256-bit secret (`crypto.randomBytes(32)`) paired with a 6-character base62 short code used as an opaque lookup key in the URL path. The QR code encodes a URL like `https://abc-xyz.trycloudflare.com/q/Xk9mQ3` — the short code is a pointer, not the secret itself, so it never leaks through browser history, `Referer` headers, or Cloudflare edge logs.
+
+Every **60 seconds**, the server automatically rotates to a fresh token. The previous token remains valid for a **90-second grace period** to handle the race where you scan right as rotation happens — after that, it's dead. Each token is **single-use**: the moment a phone successfully scans it, the token is atomically consumed and a new one is immediately generated for the desktop display.
+
+#### Security Design
+
+The design is informed by ["Demystifying the (In)Security of QR Code-based Login"](https://www.usenix.org/conference/usenixsecurity25/presentation/zhang-xin) (USENIX Security 2025), which found 47 of the top-100 websites vulnerable to QR auth attacks due to 6 critical design flaws across 42 CVEs. Codeman addresses all six:
+
+| USENIX Flaw | Mitigation |
+|-------------|------------|
+| **Flaw-1**: Missing single-use enforcement | Token atomically consumed on first scan — replays always fail |
+| **Flaw-2**: Long-lived tokens | 60s TTL with 90s grace, auto-rotation via timer |
+| **Flaw-3**: Predictable token generation | `crypto.randomBytes(32)` — 256-bit entropy. Short codes use rejection sampling to eliminate modulo bias |
+| **Flaw-4**: Client-side token generation | Server-side only — tokens never leave the server until embedded in the QR |
+| **Flaw-5**: Missing status notification | Desktop toast: *"Device [IP] authenticated via QR (Safari). Not you? [Revoke]"* — real-time QRLjacking detection |
+| **Flaw-6**: Inadequate session binding | IP + User-Agent stored for audit. Manual session revocation via API. HttpOnly + Secure + SameSite=lax cookies |
+
+#### Timing-Safe Lookup
+
+Short codes are stored in a `Map<shortCode, TokenRecord>`. Validation uses `Map.get()` — a hash-based O(1) lookup that reveals nothing about the target string through response timing. There is no character-by-character string comparison anywhere in the hot path, eliminating timing side-channel attacks entirely.
+
+#### Rate Limiting (Dual Layer)
+
+QR auth has its own rate limiting, completely independent from password auth:
+
+- **Per-IP**: 10 failed QR attempts per IP trigger a 429 block (15-minute decay window) — separate counter from Basic Auth failures, so a fat-fingered password doesn't burn your QR budget
+- **Global**: 30 QR attempts per minute across all IPs combined — defends against distributed brute force. With 62^6 = 56.8 billion possible short codes and only ~2 valid at any time, brute force is computationally infeasible regardless
+
+#### QR Code Size Optimization
+
+The URL is kept deliberately short (`/q/` path + 6-char code = ~53-56 total characters) to target **QR Version 4** (33x33 modules) instead of Version 5 (37x37). Smaller QR codes scan faster on budget phones — modern devices read Version 4 in 100-300ms. The `/q/` prefix saves 7 bytes compared to `/qr-auth/`, which alone is the difference between QR versions.
+
+#### Desktop Experience
+
+The QR display auto-refreshes every 60 seconds via SSE with the SVG embedded directly in the event payload (~2-5KB) — no extra HTTP fetch, sub-50ms refresh. A countdown timer shows time remaining. A "Regenerate" button instantly invalidates all existing tokens and creates a fresh one (useful if you suspect the QR was photographed).
+
+When someone authenticates via QR, the desktop shows a notification toast with the device's IP and browser — if it wasn't you, one click revokes all sessions.
+
+#### Threat Coverage
+
+| Threat | Why it doesn't work |
+|--------|-------------------|
+| **QR screenshot shared** | Single-use: consumed on first scan. 60s TTL: expired before the attacker can act. Desktop notification alerts you immediately. |
+| **Replay attack** | Atomic single-use consumption + 60s TTL. Old URLs always return 401. |
+| **Cloudflare edge logs** | Short code is an opaque 6-char lookup key, not the real 256-bit token. Single-use means replaying from logs always fails. |
+| **Brute force** | 56.8 billion combinations, ~2 valid at any time, dual-layer rate limiting blocks well before statistical feasibility. |
+| **QRLjacking** | 60s rotation forces real-time relay. Desktop toast provides instant detection. Self-hosted single-user context makes phishing implausible. |
+| **Timing attack** | Hash-based Map lookup — no string comparison timing leak. |
+| **Session cookie theft** | HttpOnly + Secure + SameSite=lax + 24h TTL. Manual revocation at `POST /api/auth/revoke`. |
+
+#### How It Compares
+
+| Platform | Model | Comparison |
+|----------|-------|------------|
+| **Discord** | Long-lived token, no confirmation, [repeatedly exploited](https://owasp.org/www-community/attacks/Qrljacking) | Codeman: single-use + TTL + notification |
+| **WhatsApp Web** | Phone confirms "Link device?", ~60s rotation | Comparable rotation; WhatsApp adds explicit confirmation (acceptable tradeoff for single-user) |
+| **Signal** | Ephemeral public key, E2E encrypted channel | Stronger crypto, but [exploited by Russian state actors in 2025](https://cloud.google.com/blog/topics/threat-intelligence/russia-targeting-signal-messenger) via social engineering despite it |
+
+> Full design rationale, security analysis, and implementation details: [`docs/qr-auth-plan.md`](docs/qr-auth-plan.md)
+
 ---
 
 ## SSH Alternative (`sc`)
@@ -376,6 +515,23 @@ See [CLAUDE.md](./CLAUDE.md) for full documentation.
 
 ---
 
+## Codebase Quality
+
+The codebase went through a comprehensive 7-phase refactoring that eliminated god objects, centralized configuration, and established modular architecture:
+
+| Phase | What changed | Impact |
+|-------|-------------|--------|
+| **Performance** | Cached endpoints, SSE adaptive batching, buffer chunking | Sub-16ms terminal latency |
+| **Route extraction** | `server.ts` split into 13 domain route modules + auth middleware + port interfaces | **−60%** server.ts LOC (6,736 → 2,697) |
+| **Domain splitting** | `types.ts` → 14 domain files, `ralph-tracker` → 7 files, `respawn-controller` → 5 files, `session` → 6 files | No more god files |
+| **Frontend modules** | `app.js` → 9 extracted modules (constants, mobile, voice, notifications, keyboard, CJK input, API, Ralph wizard, subagent windows) | **−24%** app.js LOC (15.2K → 11.5K) |
+| **Config consolidation** | ~70 scattered magic numbers → 9 domain-focused config files | Zero cross-file duplicates |
+| **Test infrastructure** | Shared mock library, 12 route test files, consolidated MockSession | Testable route handlers via `app.inject()` |
+
+Full details: [`docs/code-structure-findings.md`](docs/code-structure-findings.md)
+
+---
+
 ## Published Packages
 
 ### [`xterm-zerolag-input`](https://www.npmjs.com/package/xterm-zerolag-input)

config/eslint.config.js

@@ -22,8 +22,7 @@ export default tseslint.config(
       'src/web/public/vendor/**',
       'src/web/public/app.js',
       'scripts/**/*.mjs',
-      'tools/**',
-      'remotion/**',
+      'scripts/remotion/**',
     ],
   }
 );

config/vitest.config.ts

@@ -1,7 +1,11 @@
+import { resolve } from 'node:path';
 import { defineConfig } from 'vitest/config';
 
+const root = resolve(import.meta.dirname, '..');
+
 export default defineConfig({
   test: {
+    root,
     globals: true,
     environment: 'node',
     include: ['test/**/*.test.ts'],

docs/code-structure-findings.md

@@ -0,0 +1,977 @@
+# Code Structure & Quality Findings
+
+**Date**: 2026-02-28
+**Scope**: Full codebase analysis across 5 dimensions: frontend, backend, TypeScript, testing, and utilities/config.
+
+This document contains detailed findings for agent teams to write implementation plans and execute improvements. Each section includes severity, specific locations, and recommended fixes.
+
+---
+
+## Table of Contents
+
+1. [Critical: server.ts God Object (6,736 LOC)](#1-critical-serverts-god-object)
+2. [Critical: app.js Monolith (15,196 LOC)](#2-critical-appjs-monolith)
+3. [Critical: CleanupManager Unused Despite Existing](#3-critical-cleanupmanager-unused)
+4. [High: Duplicated Debounce/Timer Patterns](#4-high-duplicated-debouncetimer-patterns)
+5. [High: Large Domain Files Need Splitting](#5-high-large-domain-files-need-splitting)
+6. [High: types.ts God File (1,443 LOC)](#6-high-typests-god-file)
+7. [High: Zod Schemas Duplicate TypeScript Types](#7-high-zod-schemas-duplicate-typescript-types)
+8. [High: Test Coverage Gaps](#8-high-test-coverage-gaps)
+9. [High: Duplicated Test Mocks](#9-high-duplicated-test-mocks)
+10. [Medium: Hardcoded Magic Values](#10-medium-hardcoded-magic-values)
+11. [Medium: Frontend Global State Monolith](#11-medium-frontend-global-state-monolith)
+12. [Medium: Frontend Code Duplication](#12-medium-frontend-code-duplication)
+13. [Medium: Inconsistent Logging](#13-medium-inconsistent-logging)
+14. [Medium: Utils Barrel Export Gaps](#14-medium-utils-barrel-export-gaps)
+15. [Medium: Non-Null Assertion Risks](#15-medium-non-null-assertion-risks)
+16. [Low: Dead Utility Functions](#16-low-dead-utility-functions)
+17. [Low: No Dependency Injection for File I/O](#17-low-no-dependency-injection-for-file-io)
+18. [Scorecard & Prioritized Roadmap](#18-scorecard--prioritized-roadmap)
+
+---
+
+## 1. Critical: server.ts God Object
+
+**File**: `src/web/server.ts` (6,736 lines)
+**Severity**: CRITICAL
+**Impact**: Hardest file to maintain, test, and extend. Imports 38 modules.
+
+### Problem
+
+The `WebServer` class handles everything: HTTP routing (~110 routes), authentication, SSE broadcasting, terminal data batching, state persistence, session lifecycle, respawn orchestration, file serving, tunnel management, plan orchestration, and subagent coordination.
+
+**Key metrics**:
+- 40+ private properties (Maps, timers, caches)
+- 70+ methods
+- `setupRoutes()` is 2,000+ LOC of inline route handlers
+- Zero test coverage
+
+### Current Structure (Bad)
+
+```
+WebServer class (6,736 LOC)
+├── Auth session management (lines 469, 668-698)
+├── SSE client management (lines 407-408, 5843-5880)
+├── Terminal data batching (lines 414-416, 5909-5966)
+├── Task update batching (line 426, 5995-6028)
+├── State persistence batching (lines 429-430, 6028-6061)
+├── Respawn lifecycle (lines 445-451, 5425-5534)
+├── Session cleanup (lines 4769-4961)
+├── Listener setup (lines 544-643)
+└── setupRoutes() (lines 645+, 2000+ LOC)
+    ├── /api/sessions/* (30+ routes inline)
+    ├── /api/respawn/* (7 routes inline)
+    ├── /api/subagents/* (7 routes inline)
+    ├── /api/plan/* (5 routes inline)
+    ├── /api/push/* (4 routes inline)
+    └── ... 60+ more inline
+```
+
+### Recommended Structure
+
+```
+src/web/
+├── server.ts              (~500 LOC - HTTP setup, route registration only)
+├── routes/
+│   ├── session-routes.ts  (session CRUD, input, resize)
+│   ├── respawn-routes.ts  (respawn control endpoints)
+│   ├── subagent-routes.ts (background agent tracking)
+│   ├── plan-routes.ts     (plan generation & management)
+│   ├── push-routes.ts     (web push subscriptions)
+│   ├── mux-routes.ts      (tmux management)
+│   ├── case-routes.ts     (case management)
+│   ├── file-routes.ts     (file browsing/serving)
+│   └── system-routes.ts   (status, stats, config, settings)
+├── middleware/
+│   ├── auth.ts            (Basic Auth + session cookies)
+│   └── error-handler.ts   (centralized error responses)
+└── services/
+    ├── sse-manager.ts     (SSE client + broadcast)
+    ├── terminal-batcher.ts (60fps terminal batching)
+    └── session-lifecycle.ts (listener setup/teardown)
+```
+
+### Duplication in server.ts
+
+**Error response pattern** repeated 189 times:
+```typescript
+return createErrorResponse(ApiErrorCode.NOT_FOUND, 'Session not found');
+```
+
+**Fix**: Extract `findSessionOrFail()` middleware:
+```typescript
+const findSessionOrFail = (sessionId: string) => {
+  const session = this.sessions.get(sessionId);
+  if (!session) throw new NotFoundError('Session not found');
+  return session;
+};
+```
+
+**Event listener setup** copy-pasted for subagent watcher, image watcher, and team watcher (lines 544-643). Same attach/detach pattern duplicated 3 times.
+
+---
+
+## 2. Critical: app.js Monolith
+
+**File**: `src/web/public/app.js` (15,196 lines)
+**Severity**: CRITICAL
+**Impact**: Untestable, hard to navigate, tightly coupled systems.
+
+### Extractable Modules (by priority)
+
+| Module | Lines | Current Location | Impact |
+|--------|-------|------------------|--------|
+| Mobile handlers (MobileDetection, KeyboardHandler, SwipeHandler) | ~300 | lines 168-620 | High |
+| Voice input (DeepgramProvider, VoiceInput) | ~830 | lines 631-1471 | High |
+| NotificationManager | ~450 | lines 2218-2663 | High |
+| xterm-zerolag-input (inlined copy from packages/) | ~400 | lines 1756-2153 | High |
+| KeyboardAccessoryBar | ~195 | lines 1480-1680 | Medium |
+| FocusTrap | ~60 | lines 1690-1748 | Medium |
+
+### CodemanApp Class (12,000+ LOC)
+
+The main `CodemanApp` class starting at line 2665 has:
+- **60+ Maps/Sets** in the constructor (lines 2667-2805)
+- **18 Map instances** with complex cross-references (subagents, parents, teams, windows)
+- **10+ monolithic methods** exceeding 100 lines each
+
+**Largest methods**:
+| Method | Lines | Size |
+|--------|-------|------|
+| `renderAppSettings()` | 14400-14700 | ~300 LOC |
+| `selectSession()` | 6028-6250 | ~220 LOC |
+| `batchTerminalWrite()` | 7482-7700 | ~200 LOC |
+| `renderSessionTabs()` | 5814-6000 | ~180 LOC |
+| `openSubagentWindow()` | 11927-12100 | ~170 LOC |
+| `handleInit()` | 5183-5350 | ~170 LOC |
+
+### Recommended Split
+
+```
+src/web/public/
+├── app.js                 (~4000 LOC - core app, session mgmt, SSE)
+├── mobile.js              (~300 LOC - MobileDetection, KeyboardHandler, SwipeHandler)
+├── voice.js               (~830 LOC - DeepgramProvider, VoiceInput)
+├── notifications.js       (~450 LOC - NotificationManager)
+├── keyboard-accessory.js  (~200 LOC - KeyboardAccessoryBar)
+├── api-client.js          (~100 LOC - fetch wrapper with error handling)
+└── config.js              (~50 LOC - magic numbers, z-index layers)
+```
+
+---
+
+## 3. Critical: CleanupManager Unused
+
+**File**: `src/utils/cleanup-manager.ts` (320 lines)
+**Severity**: CRITICAL
+**Impact**: Memory leak risk. Well-designed utility exists but is never used. Every file manages cleanup manually.
+
+### Current State
+
+`CleanupManager` is exported from the utils barrel but has **0 instantiations** in production code. Instead, every file implements manual cleanup:
+
+**respawn-controller.ts** (worst offender):
+```typescript
+// 11 timer properties, manually cleared in stop()
+private stepTimer: NodeJS.Timeout | null = null;
+private completionConfirmTimer: NodeJS.Timeout | null = null;
+private noOutputTimer: NodeJS.Timeout | null = null;
+// ... 8 more
+
+stop() {
+  if (this.stepTimer) clearTimeout(this.stepTimer);
+  if (this.completionConfirmTimer) clearTimeout(this.completionConfirmTimer);
+  // ... 9 more clearTimeout/clearInterval calls
+}
+```
+
+**Files that should use CleanupManager**:
+| File | Timer/Listener Count | Current Cleanup |
+|------|---------------------|-----------------|
+| `respawn-controller.ts` | 11 timers + intervals | 11 manual clearTimeout/clearInterval |
+| `web/server.ts` | 6+ timers, debounce map | Manual in stop(), some may leak |
+| `state-store.ts` | 2 debounce timers | Manual clearTimeout |
+| `push-store.ts` | 1 save timer | Manual clearTimeout |
+| `subagent-watcher.ts` | debounce map + watchers | Manual clear + close |
+| `ralph-tracker.ts` | 3 debounce timers | Manual clear |
+| `bash-tool-parser.ts` | 1 debounce timer | Manual clear |
+| `image-watcher.ts` | 1 debounce map | Manual clear |
+
+### Fix
+
+Migrate all timer management to use `CleanupManager`. Example for respawn-controller.ts:
+
+```typescript
+// Before: 11 fields + 11 clearTimeout calls
+private stepTimer: NodeJS.Timeout | null = null;
+// ...
+
+// After: 1 field, auto-cleanup
+private cleanup = new CleanupManager();
+
+startStep() {
+  this.cleanup.setTimeout(() => { ... }, 5000, 'step');
+}
+
+stop() {
+  this.cleanup.dispose(); // Clears everything
+}
+```
+
+---
+
+## 4. High: Duplicated Debounce/Timer Patterns
+
+**Severity**: HIGH
+**Impact**: 8+ files implement debounce independently. Bug fixes need to be applied everywhere.
+
+### Pattern Inventory
+
+```typescript
+// Pattern 1: Manual timer ref (used in 6 files)
+private saveTimer: NodeJS.Timeout | null = null;
+debouncedSave() {
+  if (this.saveTimer) clearTimeout(this.saveTimer);
+  this.saveTimer = setTimeout(() => this.save(), 500);
+}
+
+// Pattern 2: Timer Map (used in 3 files)
+private fileDebouncers = new Map<string, NodeJS.Timeout>();
+debounce(key: string) {
+  const existing = this.fileDebouncers.get(key);
+  if (existing) clearTimeout(existing);
+  this.fileDebouncers.set(key, setTimeout(() => { ... }, 100));
+}
+
+// Pattern 3: State flag (used in 2 files)
+private isSaving = false;
+```
+
+### Locations
+
+| File | Debounce Vars | Delay (ms) |
+|------|---------------|------------|
+| `state-store.ts` | `saveTimeout`, `ralphStateSaveTimeout` | 500 |
+| `push-store.ts` | `saveTimer` | 500 |
+| `web/server.ts` | `persistDebounceTimers` (Map) | 500 |
+| `subagent-watcher.ts` | `fileDebouncers` (Map) | 100 |
+| `ralph-tracker.ts` | 3 debounce timers | 50, 30000 |
+| `bash-tool-parser.ts` | `EVENT_DEBOUNCE_MS` | 50 |
+| `image-watcher.ts` | debounce map | 200 |
+| `respawn-controller.ts` | 11 timer fields | various |
+
+### Fix
+
+Create a `Debouncer` utility:
+
+```typescript
+// src/utils/debouncer.ts
+export class Debouncer {
+  private timer: NodeJS.Timeout | null = null;
+
+  constructor(private readonly delayMs: number) {}
+
+  run(fn: () => void): void {
+    if (this.timer) clearTimeout(this.timer);
+    this.timer = setTimeout(fn, this.delayMs);
+  }
+
+  cancel(): void {
+    if (this.timer) clearTimeout(this.timer);
+    this.timer = null;
+  }
+}
+
+// Usage:
+private saveDeb = new Debouncer(500);
+this.saveDeb.run(() => this.save());
+// cleanup: this.saveDeb.cancel();
+```
+
+---
+
+## 5. High: Large Domain Files Need Splitting
+
+**Severity**: HIGH
+**Impact**: Complex state machines spanning 3,000+ lines are hard to understand and test.
+
+### ralph-tracker.ts (3,905 LOC)
+
+**5 responsibilities mixed**:
+1. Output Parsing (~900 LOC) - Line-by-line parsing, state extraction
+2. Todo Management (~700 LOC) - Parsing, dedup, expiry
+3. Plan Tracking (~800 LOC) - Enhanced plan tasks, checkpoints
+4. Circuit Breaker (~400 LOC) - State machine for stuck detection
+5. File Watching (~300 LOC) - Monitor external state files
+
+**Recommended split**:
+```
+ralph-tracker.ts          (core output parsing, ~1200 LOC)
+ralph-todo-manager.ts     (todo parsing + management, ~700 LOC)
+ralph-plan-tracker.ts     (plan tasks + checkpoints, ~800 LOC)
+ralph-circuit-breaker.ts  (circuit breaker logic, ~400 LOC)
+```
+
+### respawn-controller.ts (3,611 LOC)
+
+**6 responsibilities mixed**:
+1. State Machine (~1,000 LOC) - 6+ states, transitions
+2. Idle Detection (~800 LOC) - 5 layers + multi-signal combining
+3. AI Checkers (~600 LOC) - Idle + plan checkers integration
+4. Health Scoring (~500 LOC) - Metrics, circuit breaker, scoring
+5. Action Logging (~300 LOC) - Timeline, detection status
+6. Stuck-State Detection (~250 LOC) - Timeout tracking
+
+**Recommended split**:
+```
+respawn-controller.ts         (state machine core, ~1000 LOC)
+respawn-idle-detection.ts     (all 5 idle detection layers, ~800 LOC)
+respawn-health-scorer.ts      (metrics & health scoring, ~500 LOC)
+```
+
+### session.ts (2,418 LOC)
+
+**8 responsibilities mixed**:
+1. PTY Management (~600 LOC)
+2. Terminal I/O (~400 LOC)
+3. Token Tracking (~200 LOC)
+4. Task Tracking (~250 LOC)
+5. Ralph Integration (~200 LOC)
+6. Auto-Clear/Compact (~300 LOC)
+7. Image Watching (~100 LOC)
+8. CLI Detection (~150 LOC)
+
+**Recommended split**:
+```
+session.ts               (PTY + terminal I/O core, ~1000 LOC)
+session-tracking.ts      (token + task + Ralph, ~500 LOC)
+session-auto-ops.ts      (auto-clear/compact + image, ~300 LOC)
+```
+
+---
+
+## 6. High: types.ts God File
+
+**File**: `src/types.ts` (1,443 lines, 72 exported definitions)
+**Severity**: HIGH
+**Impact**: Every file imports from types.ts. Hard to find relevant types.
+
+### Current Contents
+
+- 46 interfaces
+- 25 types
+- 1 enum (ApiErrorCode)
+- 9 factory functions (createInitialState, etc.)
+
+### Recommended Split
+
+```
+src/types/
+├── index.ts          (barrel export - transparent migration)
+├── session.ts        (SessionState, SessionConfig, SessionMode, SessionColor)
+├── task.ts           (TaskState, TaskDefinition, TaskStatus)
+├── respawn.ts        (RespawnConfig, RespawnState, CircuitBreakerStatus)
+├── ralph.ts          (RalphLoopState, RalphTrackerState, RalphTodoItem)
+├── api.ts            (ApiResponse, ApiErrorCode, HookEventType, all route types)
+├── lifecycle.ts      (LifecycleEventType, LifecycleEntry)
+└── common.ts         (Disposable, BufferConfig, CleanupResourceType)
+```
+
+The barrel export makes this a transparent refactor - existing `import from './types'` continues to work.
+
+---
+
+## 7. High: Zod Schemas Duplicate TypeScript Types
+
+**File**: `src/web/schemas.ts` (508 lines)
+**Severity**: HIGH
+**Impact**: When a type changes, the Zod schema must be manually updated too. Source of bugs.
+
+### Problem
+
+Zod schemas manually duplicate TypeScript interfaces. **Zero `z.infer` usage found.**
+
+```typescript
+// types.ts (manual interface)
+export interface CreateSessionRequest {
+  workingDir?: string;
+  mode?: SessionMode;
+  name?: string;
+}
+
+// schemas.ts (manual Zod schema - duplicated!)
+export const CreateSessionSchema = z.object({
+  workingDir: safePathSchema.optional(),
+  mode: z.enum(['claude', 'shell', 'opencode']).optional(),
+  name: z.string().max(100).optional(),
+});
+```
+
+### Fix
+
+Use `z.infer` to derive TypeScript types from Zod schemas (single source of truth):
+
+```typescript
+// schemas.ts
+export const CreateSessionSchema = z.object({
+  workingDir: safePathSchema.optional(),
+  mode: z.enum(['claude', 'shell', 'opencode']).optional(),
+  name: z.string().max(100).optional(),
+});
+
+// types.ts (auto-derived)
+export type CreateSessionRequest = z.infer<typeof CreateSessionSchema>;
+```
+
+**Affected schemas** (~10):
+- CreateSessionSchema
+- RunPromptSchema
+- ResizeSchema
+- CreateCaseSchema
+- QuickStartSchema
+- HookEventSchema
+- RespawnConfigSchema
+- ConfigUpdateSchema
+- SettingsUpdateSchema
+
+---
+
+## 8. High: Test Coverage Gaps
+
+**Severity**: HIGH
+**Impact**: Critical code paths untested. Regressions go unnoticed.
+
+### Untested Source Files
+
+| File | Lines | Risk |
+|------|-------|------|
+| `src/web/server.ts` | 6,736 | CRITICAL - Core REST API, 280+ routes |
+| `src/plan-orchestrator.ts` | ~500 | HIGH - Multi-agent plan generation |
+| `src/tunnel-manager.ts` | ~200 | MEDIUM - Cloudflare tunnel |
+| `src/session-lifecycle-log.ts` | ~150 | MEDIUM - JSONL audit log |
+| `src/ai-plan-checker.ts` | ~300 | MEDIUM - Plan completion detection |
+| `src/templates/claude-md.ts` | ~200 | LOW - CLAUDE.md generation |
+| `src/utils/claude-cli-resolver.ts` | ~100 | LOW - CLI path resolution |
+| `src/utils/opencode-cli-resolver.ts` | ~100 | LOW - OpenCode CLI support |
+| `src/utils/regex-patterns.ts` | ~100 | LOW - Used everywhere! |
+| `src/utils/token-validation.ts` | ~50 | LOW - Token counting |
+
+### Test Quality Issues
+
+**10 "not.toThrow()" tests without behavior verification**:
+```typescript
+// BAD: Only checks it doesn't crash
+expect(() => tracker.processMessage(null)).not.toThrow();
+
+// GOOD: Also verify defensive behavior
+expect(() => tracker.processMessage(null)).not.toThrow();
+expect(tracker.getAllTasks().size).toBe(0);
+```
+
+Locations:
+- `task-tracker.test.ts` - 5 instances
+- `image-watcher.test.ts` - 1 instance
+- `task-queue.test.ts` - 1 instance
+- Others scattered
+
+---
+
+## 9. High: Duplicated Test Mocks
+
+**Severity**: HIGH
+**Impact**: Mock changes need updating in 4 places. Inconsistent mock behavior.
+
+### MockSession Defined 4 Times
+
+| File | Usage |
+|------|-------|
+| `test/respawn-controller.test.ts` | Full mock with event emitter |
+| `test/session-manager.test.ts` | Simpler mock |
+| `test/respawn-team-awareness.test.ts` | Copy of respawn-controller mock |
+| `test/respawn-test-utils.ts` | **Comprehensive mock - UNUSED!** |
+
+### MockStateStore Defined 2 Times
+
+| File | Usage |
+|------|-------|
+| `test/session-manager.test.ts` | Basic mock |
+| `test/ralph-loop.test.ts` | Separate implementation |
+
+### Unused Test Utilities
+
+`test/respawn-test-utils.ts` exports these utilities that **no test file imports**:
+- `createTimeController()` - Abstraction over vitest fake timers
+- `MockAiIdleChecker` - Fully mocked AI idle checker
+- `MockAiPlanChecker` - Fully mocked plan checker
+- Factory functions for pre-configured controllers
+
+### Fix
+
+Create `test/mocks/` directory:
+```
+test/
+├── mocks/
+│   ├── mock-session.ts      (single MockSession, used everywhere)
+│   ├── mock-state-store.ts  (single MockStateStore)
+│   └── index.ts             (barrel export)
+├── utils/
+│   └── time-controller.ts   (from respawn-test-utils.ts)
+└── ... test files
+```
+
+---
+
+## 10. Medium: Hardcoded Magic Values
+
+**Severity**: MEDIUM
+**Impact**: Hard to tune, inconsistent when same value appears in multiple places.
+
+### Already Centralized (Good)
+
+- `src/config/buffer-limits.ts` - All buffer sizes
+- `src/config/map-limits.ts` - All collection limits
+
+### NOT Centralized (40+ values scattered)
+
+**In server.ts** (lines 145-194):
+```typescript
+const TASK_UPDATE_BATCH_INTERVAL = 100;
+const STATE_UPDATE_DEBOUNCE_INTERVAL = 500;
+const SESSIONS_LIST_CACHE_TTL = 1000;
+const SCHEDULED_CLEANUP_INTERVAL = 5 * 60 * 1000;
+const SSE_HEALTH_CHECK_INTERVAL = 30 * 1000;
+const MAX_TERMINAL_COLS = 500;
+const MAX_TERMINAL_ROWS = 200;
+const AUTH_SESSION_TTL_MS = 24 * 60 * 60 * 1000;
+const MAX_AUTH_SESSIONS = 100;
+const AUTH_FAILURE_WINDOW_MS = 15 * 60 * 1000;
+const STATS_COLLECTION_INTERVAL_MS = 2000;
+const MAX_INPUT_LENGTH = 64 * 1024;
+```
+
+**In hooks-config.ts**: `timeout: 10000` hardcoded 6 times.
+
+**In respawn-controller.ts** (lines 538-565): 10 timing constants.
+
+**In utils**: `EXEC_TIMEOUT_MS = 5000` duplicated in both `claude-cli-resolver.ts` and `opencode-cli-resolver.ts`.
+
+**In app.js**:
+```javascript
+// line 27: 600000 - stuck detection threshold
+// line 24: 5000 - default scrollback
+// lines 34-35: 128*1024, 256*1024 - chunk sizes
+// lines 152-155: 150, 100 - keyboard detection thresholds
+// lines 573-575: 80, 300, 100 - swipe detection params
+```
+
+### Fix
+
+Create additional config files:
+```
+src/config/
+├── buffer-limits.ts     (existing)
+├── map-limits.ts        (existing)
+├── server-config.ts     (NEW - web server intervals, auth, caching)
+├── timing-config.ts     (NEW - debounce delays, check intervals)
+└── terminal-config.ts   (NEW - max cols/rows, batch intervals)
+```
+
+---
+
+## 11. Medium: Frontend Global State Monolith
+
+**Severity**: MEDIUM
+**Impact**: All state in single CodemanApp class. Tight coupling between unrelated systems.
+
+### 60+ State Variables in CodemanApp Constructor (lines 2667-2805)
+
+```javascript
+this.sessions = new Map();             // Session data
+this.subagents = new Map();            // Agent tracking
+this.subagentActivity = new Map();     // Tool call tracking
+this.subagentToolResults = new Map();  // Result caching
+this.subagentParentMap = new Map();    // Agent-to-session mapping
+this.teams = new Map();                // Team tracking
+this.teamTasks = new Map();           // Team task state
+this.planSubagents = new Map();        // Plan agent tracking
+this.pendingWrites = [];               // Terminal write queue
+this.terminalBufferCache = new Map();  // Buffer caching (unbounded!)
+this.projectInsights = new Map();      // Bash tool insights
+// ... 40+ more
+```
+
+### Problems
+
+1. **18 Map instances** with complex cross-references (no garbage collection strategy)
+2. **No domain separation**: Session, subagent, notification, UI, and network state mixed
+3. **Implicit dependencies**: `selectSession()` requires 5+ Maps to be in consistent state
+4. **`terminalBufferCache`** has no max size - can grow unbounded with many sessions
+
+### Recommended Domain Split
+
+```javascript
+// Instead of 60+ flat properties:
+class SessionState {
+  sessions = new Map();
+  sessionOrder = [];
+  terminalBuffers = new Map();
+  tabAlerts = new Map();
+}
+
+class SubagentState {
+  subagents = new Map();
+  activity = new Map();
+  parentMap = new Map();
+  windows = new Map();
+  minimized = new Map();
+}
+
+class TeamState {
+  teams = new Map();
+  tasks = new Map();
+  teammates = new Map();
+}
+
+class UIState {
+  activeSessionId = null;
+  draggedTabId = null;
+  isLoadingBuffer = false;
+}
+```
+
+---
+
+## 12. Medium: Frontend Code Duplication
+
+**Severity**: MEDIUM
+**Impact**: Repeated patterns increase maintenance burden and inconsistency risk.
+
+### Duplicated Patterns
+
+**API fetch calls** (~50 instances):
+```javascript
+// Repeated everywhere:
+fetch(`/api/sessions/${sessionId}/...`, {
+  method: 'POST',
+  headers: { 'Content-Type': 'application/json' },
+  body: JSON.stringify({...})
+}).catch(() => {})
+```
+**Fix**: Extract `ApiClient` class.
+
+**`innerHTML` usage** (104 instances):
+- Mix of template strings, createElement chains, and direct innerHTML
+- Some with manual XSS escaping (`text.replace(/</g, '&lt;')`), some without
+- No consistent DOM creation pattern
+
+**`typeof app !== 'undefined'` checks** (20+ instances):
+- Lines 458, 467, 481, 614, 617, 1549, etc.
+- **Fix**: Ensure `app` is always defined as global singleton.
+
+**Element visibility toggling** (212+ occurrences):
+```javascript
+element.classList.add('active')
+element.classList.remove('active')
+```
+**Fix**: Create `toggleClass(el, className, condition)` utility.
+
+### Event Listener Issues
+
+- **152 `addEventListener` calls** with fragile cleanup
+- **Mix of inline (`onclick="app.method()"`) and addEventListener** - hard to track
+- **Element cache (`_elemCache`) never invalidated** if DOM elements are recreated (line 2808)
+- **Tab drag-and-drop listeners** may not clean up if user switches tabs mid-drag
+
+---
+
+## 13. Medium: Inconsistent Logging
+
+**Severity**: MEDIUM
+**Impact**: Hard to debug in production. Can't filter by severity or component.
+
+### Current State
+
+- **345 console calls** across source files
+- **No structured logging** - all `console.log/error` directly
+- **No log levels** (DEBUG, INFO, WARN, ERROR)
+
+### Inconsistent Prefixes
+
+```typescript
+// Some files use brackets:
+console.log('[Session] Starting interactive...');
+console.log('[RalphLoop] Task assigned...');
+console.log('[TunnelManager] Tunnel started');
+
+// Others use no prefix:
+console.error('Failed to spawn PTY:', err);
+console.log('Server listening on port', port);
+```
+
+### Positive: CleanupManager Has Debug Mode
+
+`src/utils/cleanup-manager.ts` has a `debugMode` flag for conditional debug logging - good pattern not replicated elsewhere.
+
+### Fix
+
+Either:
+1. Enforce consistent `[ComponentName]` prefixes via lint rule
+2. Create lightweight logger abstraction (not a heavy framework)
+
+---
+
+## 14. Medium: Utils Barrel Export Gaps
+
+**File**: `src/utils/index.ts`
+**Severity**: MEDIUM
+**Impact**: Forces deep imports, unclear public API.
+
+### Missing Exports
+
+These functions are defined but NOT exported from the barrel:
+- `createAnsiPatternFull()` and `createAnsiPatternSimple()` (factory functions from `regex-patterns.ts`)
+- `SAFE_PATH_PATTERN` (from `regex-patterns.ts`)
+- `validateTokenCounts()` and `validateTokensAndCost()` (from `token-validation.ts`)
+- `isSimilar()`, `isSimilarByDistance()`, `levenshteinDistance()`, `normalizePhrase()` (from `string-similarity.ts` - though some are dead code, see finding #16)
+
+### Deep Import Anti-Pattern (16 instances)
+
+Some files bypass the barrel unnecessarily:
+```typescript
+// Could use barrel:
+import { BufferAccumulator } from './utils/buffer-accumulator.js';
+import { LRUMap } from './utils/lru-map.js';
+
+// Must deep import (not in barrel):
+import { SAFE_PATH_PATTERN } from './utils/regex-patterns.js';
+```
+
+### Fix
+
+Add missing exports to `src/utils/index.ts` and update import sites.
+
+---
+
+## 15. Medium: Non-Null Assertion Risks
+
+**Severity**: MEDIUM
+**Impact**: Runtime crashes if assumptions violated. 37 instances found.
+
+### Distribution
+
+| File | Count | Risk Level |
+|------|-------|------------|
+| `src/web/server.ts` | 10 | Low (auth flow verified) |
+| `src/session.ts` | 6 | **High** (mux/terminal refs) |
+| `src/respawn-controller.ts` | 4 | Low (config validated) |
+| `src/lru-map.ts` | 3 | Low (checked lookups) |
+| `src/subagent-watcher.ts` | 2 | Low (pending tool calls) |
+| Others | 12 | Low |
+
+### High-Risk Examples (session.ts)
+
+```typescript
+// Line 915 - _mux could be null if startInteractive called during cleanup
+`[Session] Starting interactive (with ${this._mux!.backend})`
+
+// Line 954 - _muxSession could be null in race condition
+this._muxSession!.muxName
+```
+
+### Fix
+
+Add null guards before assertions, or document invariants:
+```typescript
+// Before:
+this._mux!.backend
+
+// After:
+if (!this._mux) throw new Error('Invariant: _mux must be initialized before startInteractive');
+this._mux.backend
+```
+
+### Positive Notes
+
+- **0 instances of `as any`**
+- **0 instances of `@ts-ignore` or `@ts-expect-error`**
+- TypeScript overall score: 8.5/10
+
+---
+
+## 16. Low: Dead Utility Functions
+
+**File**: `src/utils/string-similarity.ts`
+**Severity**: LOW
+**Impact**: Code clutter, confusion about what's actually used.
+
+### Unused Functions
+
+These are defined and exported but **never imported anywhere**:
+- `isSimilar(a, b, threshold)` - similarity check with threshold
+- `isSimilarByDistance(a, b, maxDistance)` - Levenshtein-based check
+- `levenshteinDistance(a, b)` - raw edit distance
+- `normalizePhrase(phrase)` - phrase normalization
+
+### Actually Used
+
+Only these are imported from the barrel:
+- `stringSimilarity()` - used in ralph-tracker.ts
+- `fuzzyPhraseMatch()` - used in ralph-tracker.ts
+- `todoContentHash()` - used in ralph-tracker.ts
+
+### Fix
+
+Delete unused functions or mark as `@internal` if kept for future use.
+
+---
+
+## 17. Low: No Dependency Injection for File I/O
+
+**Severity**: LOW (practical impact limited at current scale)
+**Impact**: Can't mock filesystem for unit tests. 68+ hard-coded filesystem calls.
+
+### Examples
+
+```typescript
+// state-store.ts - directly imports and uses fs
+import { readFileSync, writeFileSync, existsSync, mkdirSync } from 'node:fs';
+
+// push-store.ts - hard-coded paths
+const KEYS_FILE = join(DATA_DIR, 'push-keys.json');
+const SUBS_FILE = join(DATA_DIR, 'push-subscriptions.json');
+
+// ai-checker-base.ts - direct execSync
+execSync(`tmux kill-session -t "${this.checkMuxName}"`, { timeout: 3000 });
+```
+
+### Why This Is Lower Priority
+
+- The codebase uses integration tests (spawning real processes/tmux sessions) rather than unit tests
+- Most filesystem operations are in infrastructure code, not business logic
+- Adding DI would be a large refactor with limited near-term benefit
+
+---
+
+## 18. Scorecard & Prioritized Roadmap
+
+### Overall Scores (Post-Implementation)
+
+| Category | Before | After | Notes |
+|----------|--------|-------|-------|
+| TypeScript Safety | 8.5/10 | 9/10 | 0 `any`, 0 `@ts-ignore`, Zod `z.infer` eliminates type drift |
+| Error Handling | 8/10 | 8/10 | Unchanged — already strong |
+| Async/Promise Safety | 9.5/10 | 9.5/10 | Unchanged — already strong |
+| Resource Cleanup | 7/10 | 8/10 | CleanupManager adopted in server.ts, subagent-watcher, bash-tool-parser; Debouncer in 6 files. **Gaps**: respawn-controller (10+ manual timers) and ralph-tracker (2 manual timers) not migrated |
+| Module Organization | 5/10 | 8/10 | Routes extracted (12 modules), types split (14 domain files), domain files split (ralph: 7, respawn: 5, session: 6) |
+| Test Coverage | 6/10 | 7.5/10 | Shared mock infrastructure, 12 route test files, MockSession/MockStateStore consolidated |
+| Config Centralization | 6/10 | 9/10 | 9 config files, ~65 constants centralized, 0 cross-file duplicates |
+| Frontend Architecture | 4/10 | 7/10 | 8 extracted modules (3,453 LOC), app.js reduced 24% (15.2K → 11.5K), xterm-zerolag-input vendor build |
+| Code Duplication | 5/10 | 8/10 | Debouncer utility, shared test mocks, barrel exports, config consolidation |
+
+### Implementation Phases
+
+**Phase 1 - Quick Wins (1-2 days)** ✅ COMPLETE
+1. ✅ Export missing functions from utils barrel (~30 min) — `createAnsiPatternFull`, `createAnsiPatternSimple`, `SAFE_PATH_PATTERN`, `validateTokenCounts`, `validateTokensAndCost` all now exported from `src/utils/index.ts`
+2. ✅ Delete dead utility functions (~15 min) — `isSimilar()` removed from `string-similarity.ts`; `levenshteinDistance()`, `isSimilarByDistance()`, `normalizePhrase()` made private (used internally by `fuzzyPhraseMatch`/`stringSimilarity`)
+3. ✅ Consolidate duplicated `EXEC_TIMEOUT_MS` constant (~15 min) — Created `src/config/exec-timeout.ts` as single source of truth; `claude-cli-resolver.ts`, `opencode-cli-resolver.ts`, and `tmux-manager.ts` all import from it
+4. ✅ Add `z.infer` to Zod schemas (~2 hours) — `src/web/schemas.ts` now has 36 `z.infer` type exports (lines 512-547) covering all schemas
+5. ✅ Fix 10 weak "not.toThrow()" tests (~1 hour) — All `not.toThrow()` calls now have behavior assertions: `task-tracker.test.ts` (6 instances all followed by state checks), `image-watcher.test.ts` (1 instance followed by length check), `session-manager.test.ts` (1 instance followed by count check)
+
+**Phase 2 - CleanupManager & Debounce (2-3 days)** ✅ COMPLETE
+1. ✅ Create `Debouncer` utility class (~1 hour) — Created `src/utils/debouncer.ts` with `Debouncer` and `KeyedDebouncer` classes; exported from `src/utils/index.ts`
+2. ✅ Migrate all 8 files from manual debounce to Debouncer — `state-store.ts` (2 Debouncers), `push-store.ts` (1 Debouncer), `bash-tool-parser.ts` (1 Debouncer), `image-watcher.ts` (1 KeyedDebouncer), `subagent-watcher.ts` (2 KeyedDebouncers), `server.ts` (1 KeyedDebouncer for persist timers), `ralph-tracker.ts` (2 Debouncers replacing 4 manual fields: `_todoUpdateTimer`, `_loopUpdateTimer`, `_todoUpdatePending`, `_loopUpdatePending`)
+3. ✅ Migrate respawn-controller to CleanupManager — 10 manual timer fields replaced with single `CleanupManager` instance + `timerIds` Map. `startTrackedTimer()`/`cancelTrackedTimer()` preserved as wrappers for UI countdown display and timer events. `clearTimers()` uses dispose-and-recreate pattern for state transitions.
+4. ✅ Migrate server.ts timer cleanup to CleanupManager (~2 hours) — `private cleanup = new CleanupManager()` present; terminal batch timers and pending respawn starts left as manual Maps (complex lifecycle)
+5. ✅ Migrate remaining files — `bash-tool-parser.ts` (CleanupManager ✅), `subagent-watcher.ts` (CleanupManager ✅), `ralph-tracker.ts` (Debouncer ✅)
+
+**Phase 3 - server.ts Route Extraction (3-4 days)** ✅ COMPLETE
+1. ✅ Created `src/web/routes/` with 12 domain route modules + index barrel (4,090 LOC total): session (909), system (768), ralph (533), plan (459), respawn (315), case, file, hook-event, mux, push, scheduled, team
+2. ✅ Created `src/web/middleware/auth.ts` (193 LOC) — Basic Auth, session cookies, rate limiting, security headers, CORS
+3. ✅ Created `src/web/ports/` with 7 typed port interfaces (142 LOC) — SessionPort, EventPort, RespawnPort, ConfigPort, InfraPort, AuthPort; routes declare dependencies via intersection types
+4. ✅ Created `src/web/route-helpers.ts` (154 LOC) — `findSessionOrFail()`, `formatUptime()`, `sanitizeHookData()`, `autoConfigureRalph()`
+5. ✅ Reduced `server.ts` from 6,736 → 2,697 LOC (60% reduction). Remaining LOC is justified infrastructure: session lifecycle, SSE broadcast engine, terminal batching, respawn integration, resource cleanup
+
+**Phase 4 - Domain File Splitting (2-3 days)** ✅ COMPLETE
+1. ✅ Split `types.ts` into `src/types/` directory — 14 domain files (1,469 LOC total): common, session, task, app-state, respawn, ralph, api, lifecycle, run-summary, tools, teams, push, plan + index barrel. Original `types.ts` is now a 1-line re-export
+2. ✅ Split `ralph-tracker.ts` into 7 files (exceeded plan of 4) — ralph-tracker (2,391), ralph-plan-tracker (477), ralph-status-parser (552), ralph-fix-plan-watcher (366), ralph-stall-detector (166), ralph-config (153), ralph-loop (522)
+3. ✅ Split `respawn-controller.ts` into 5 files (exceeded plan of 3) — respawn-controller (3,228), respawn-health (229), respawn-metrics (229), respawn-patterns (131), respawn-adaptive-timing (134)
+4. ✅ Split `session.ts` into 6 files (exceeded plan of 3) — session (2,168), session-manager (298), session-auto-ops (284), session-cli-builder (132), session-task-cache (101), session-lifecycle-log (114)
+
+**Phase 5 - Frontend Modularization (3-4 days)** ✅ COMPLETE
+1. ✅ Extracted `constants.js` (238 LOC) — shared constants, timing values, Z-index layers, `escapeHtml()`, `extractSyncSegments()`
+2. ✅ Extracted `mobile-handlers.js` (449 LOC) — `MobileDetection`, `KeyboardHandler`, `SwipeHandler`
+3. ✅ Extracted `voice-input.js` (853 LOC) — `DeepgramProvider`, `VoiceInput`
+4. ✅ Extracted `notification-manager.js` (445 LOC) — `NotificationManager` class (5-layer system)
+5. ✅ Extracted `keyboard-accessory.js` (279 LOC) — `KeyboardAccessoryBar`, `FocusTrap`
+6. ✅ Extracted `api-client.js` (70 LOC) — `_api()`, `_apiJson()`, `_apiPost()`, `_apiPut()`
+7. ✅ Extracted `subagent-windows.js` (1,119 LOC) — 13 subagent window methods
+8. ✅ Removed inlined xterm-zerolag-input copy → built to `vendor/xterm-zerolag-input.js` from `packages/xterm-zerolag-input/`
+9. ✅ Reduced `app.js` from ~15,200 → 11,473 LOC (24% reduction). All scripts loaded in correct dependency order in `index.html`
+
+**Phase 6 - Config Consolidation (1 day)** ✅ COMPLETE
+1. ✅ Created 6 new domain-focused config files (better than plan's 2 generic files): `server-timing.ts` (13 constants), `auth-config.ts` (5 constants), `tunnel-config.ts` (8 constants), `terminal-limits.ts` (4 constants), `ai-defaults.ts` (3 constants), `team-config.ts` (3 constants)
+2. ✅ Total: 9 config files in `src/config/`, ~65 constants centralized
+3. ✅ Eliminated all cross-file duplicates: `STATS_COLLECTION_INTERVAL_MS` (was in 2 files), `timeout: 10000` (was 6× inline in hooks-config.ts → `HOOK_TIMEOUT_MS`), AI model string (was in 5 files → `AI_CHECK_MODEL`), `MAX_TRACKED_AGENTS` (was shadowed in subagent-watcher.ts)
+4. ✅ CLAUDE.md updated with config files table, import conventions, resource limits references
+
+**Phase 7 - Test Infrastructure (2-3 days)** ✅ COMPLETE
+1. ✅ Created `test/mocks/` directory with 5 files (541 LOC): `mock-session.ts` (312), `mock-state-store.ts` (60), `mock-route-context.ts` (121), `test-helpers.ts` (37), `index.ts` (11 — barrel export)
+2. ✅ Consolidated MockSession into single shared definition — no duplicate class definitions remain (2 `vi.mock()`-based copies intentionally left in session-manager.test.ts and ralph-loop.test.ts)
+3. ✅ `respawn-test-utils.ts` converted to backward-compatibility shim — re-exports from `test/mocks/`, retains respawn-specific utilities (MockAiIdleChecker, TimeController, etc.)
+4. ✅ Created initial 3 route test files with 58 total tests: `session-routes.test.ts` (34 tests), `respawn-routes.test.ts` (13 tests), `system-routes.test.ts` (11 tests). Route test harness uses `app.inject()` — no real ports needed
+5. ✅ All 12 route modules now have dedicated test files in `test/routes/`: session, respawn, system, ralph, plan, push, team, mux, file, scheduled, hook-event, case
+
+---
+
+## Appendix: File Size Inventory (Post-Implementation)
+
+### Before vs After
+
+| File | Before | After | Change |
+|------|--------|-------|--------|
+| `src/web/server.ts` | 6,736 | 2,697 | **−60%** (routes, auth, ports extracted) |
+| `src/web/public/app.js` | 15,196 | 11,473 | **−24%** (8 modules extracted) |
+| `src/ralph-tracker.ts` | 3,905 | 2,391 | **−39%** (6 companion files extracted) |
+| `src/respawn-controller.ts` | 3,611 | 3,228 | **−11%** (4 companion files extracted) |
+| `src/session.ts` | 2,418 | 2,168 | **−10%** (5 companion files extracted) |
+| `src/types.ts` | 1,443 | 1 | **−99%** (14 domain files in `src/types/`) |
+
+### New Infrastructure Created
+
+| Directory | Files | Total LOC | Purpose |
+|-----------|-------|-----------|---------|
+| `src/web/routes/` | 13 | 4,090 | Domain route modules |
+| `src/web/ports/` | 7 | 142 | Port interfaces for DI |
+| `src/web/middleware/` | 1 | 193 | Auth middleware |
+| `src/types/` | 14 | 1,469 | Domain type files |
+| `src/config/` | 9 | ~450 | Centralized config |
+| `test/mocks/` | 5 | 541 | Shared test mocks |
+| `test/routes/` | 4 | ~500 | Route handler tests |
+
+### Extracted Frontend Modules
+
+| Module | Lines | Purpose |
+|--------|-------|---------|
+| `subagent-windows.js` | 1,119 | Subagent window management |
+| `voice-input.js` | 853 | DeepgramProvider, VoiceInput |
+| `mobile-handlers.js` | 449 | MobileDetection, KeyboardHandler, SwipeHandler |
+| `notification-manager.js` | 445 | 5-layer notification system |
+| `keyboard-accessory.js` | 279 | KeyboardAccessoryBar, FocusTrap |
+| `constants.js` | 238 | Shared constants, timing, Z-index |
+| `api-client.js` | 70 | API fetch wrapper |
+
+### What's Working Well
+
+These patterns should be **preserved, not refactored**:
+- Clean one-way dependency graph (no circular deps)
+- EventEmitter-based decoupling between domain models
+- Proper `import type` usage (19 files, consistent)
+- Utility type adoption (101 instances of Record, Partial, Omit, etc.)
+- `assertNever()` for exhaustive switch checking
+- `StaleExpirationMap` and `LRUMap` for bounded collections
+- State persistence circuit breaker pattern
+- TypeScript strict mode with all safety flags enabled
+- `CleanupManager` for centralized timer/watcher disposal
+- `Debouncer`/`KeyedDebouncer` for consistent debounce patterns
+- Port interfaces for route module dependency injection
+- `Object.assign(CodemanApp.prototype, ...)` for frontend module composition

docs/orchestrator-loop-architecture.md

@@ -0,0 +1,367 @@
+# Orchestrator Loop — Architecture & Data Flow
+
+> Technical architecture document. Not for GitHub.
+
+## System Overview
+
+```
+┌─────────────────────────────────────────────────────────────────────┐
+│                          CODEMAN WEB UI                             │
+│  ┌──────────────────────────────────────────────────────────────┐  │
+│  │  Orchestrator Dashboard                                       │  │
+│  │  [Goal Input] [Plan View] [Phase Progress] [Agent Activity]  │  │
+│  └───────────────────────────┬──────────────────────────────────┘  │
+│                               │ SSE Events                         │
+│                               ▼                                    │
+│  ┌──────────────────────────────────────────────────────────────┐  │
+│  │  Orchestrator API Routes (/api/orchestrator/*)                │  │
+│  └───────────────────────────┬──────────────────────────────────┘  │
+└───────────────────────────────┼─────────────────────────────────────┘
+                                ▼
+┌─────────────────────────────────────────────────────────────────────┐
+│                      ORCHESTRATOR LOOP                               │
+│                                                                      │
+│  ┌──────────────┐    ┌──────────────┐    ┌──────────────────────┐  │
+│  │ Orchestrator │    │ Orchestrator │    │ Orchestrator         │  │
+│  │ Planner      │    │ Loop (state  │    │ Verifier             │  │
+│  │              │    │ machine)     │    │                      │  │
+│  │ • Research   │◄──►│ • Phase mgmt │◄──►│ • Test runner        │  │
+│  │ • Plan gen   │    │ • Task queue │    │ • AI review          │  │
+│  │ • Phasing    │    │ • Event loop │    │ • Output checks      │  │
+│  └──────┬───────┘    └──────┬───────┘    └──────────┬───────────┘  │
+│         │                   │                       │              │
+│         ▼                   ▼                       ▼              │
+│  ┌──────────────────────────────────────────────────────────────┐  │
+│  │              EXISTING CODEMAN INFRASTRUCTURE                  │  │
+│  │                                                               │  │
+│  │  SessionManager ←→ Sessions ←→ PTY (Claude CLI)              │  │
+│  │       ↑                ↑              ↑                       │  │
+│  │       │                │              │                       │  │
+│  │  TaskQueue      RalphTracker    RespawnController             │  │
+│  │  StateStore     HooksConfig     TeamWatcher                   │  │
+│  │  Auto-Ops       SubagentWatcher SSE Broadcast                 │  │
+│  └──────────────────────────────────────────────────────────────┘  │
+└─────────────────────────────────────────────────────────────────────┘
+```
+
+## Data Flow: Complete Lifecycle
+
+### 1. User Submits Goal
+
+```
+User → POST /api/orchestrator/start { goal: "Build a REST API...", config: {...} }
+  → OrchestratorLoop.start(goal)
+    → state = PLANNING
+    → emit('stateChanged', 'planning')
+    → SSE: orchestrator:stateChanged
+```
+
+### 2. Planning Phase
+
+```
+OrchestratorPlanner.generatePlan(goal)
+  → PlanOrchestrator.generateDetailedPlan(goal)
+    → [Research Agent] → enriched task description
+    → [Planner Agent] → PlanItem[]
+  → groupIntoPhases(planItems)
+    → topological sort by dependencies
+    → group into layers
+    → assign team strategies
+  → OrchestratorPlan { phases: [...] }
+  → state = APPROVAL
+  → emit('planReady', plan)
+  → SSE: orchestrator:planReady
+```
+
+### 3. User Approves Plan
+
+```
+User → POST /api/orchestrator/approve
+  → OrchestratorLoop.approvePlan()
+    → state = EXECUTING
+    → executePhase(phases[0])
+```
+
+### 4. Phase Execution
+
+```
+executePhase(phase)
+  → For each task in phase:
+    → Convert to CreateTaskOptions
+    → Add to TaskQueue with completion phrase "PHASE_{N}_TASK_{M}_DONE"
+  → If phase.teamStrategy.type === 'team':
+    → Start session with AGENT_TEAMS enabled
+    → Send team orchestration prompt to lead
+  → Else:
+    → Assign tasks to available sessions (same as RalphLoop)
+
+  → Listen for task completion events:
+    → TaskQueue emits taskCompleted
+    → Check: all phase tasks done?
+      → Yes → state = VERIFYING → verifyPhase(phase)
+      → No → wait for more completions
+```
+
+### 5. Verification
+
+```
+verifyPhase(phase)
+  → OrchestratorVerifier.verify(phase, session)
+    → Run test commands via session
+    → Check file existence
+    → AI review (optional)
+  → If passed:
+    → phase.status = 'passed'
+    → emit('phaseCompleted', phase)
+    → If more phases: executePhase(nextPhase)
+    → If last phase: state = COMPLETED
+  → If failed:
+    → phase.attempts++
+    → If attempts < maxAttempts:
+      → state = REPLANNING
+      → Generate recovery tasks
+      → state = EXECUTING (retry)
+    → Else:
+      → state = FAILED
+      → emit('phaseFailed', phase, reason)
+```
+
+### 6. Context Management Between Phases
+
+```
+After phase completion:
+  → If config.compactBetweenPhases:
+    → session.sendInput('/compact')
+    → Wait for compact to complete
+  → If config.respawnBetweenMilestones && phase is a milestone:
+    → Save orchestrator state to StateStore
+    → Respawn session (kill + recreate)
+    → Send resume prompt with phase context
+```
+
+## File Layout
+
+```
+src/
+├── orchestrator-loop.ts          # Main state machine (~400 lines)
+├── orchestrator-planner.ts       # Plan generation + phase grouping (~300 lines)
+├── orchestrator-verifier.ts      # Phase verification (~200 lines)
+├── types/
+│   └── orchestrator.ts           # All orchestrator types (~150 lines)
+├── prompts/
+│   └── orchestrator.ts           # Prompt templates (~200 lines)
+├── web/
+│   ├── routes/
+│   │   └── orchestrator-routes.ts  # API endpoints (~250 lines)
+│   └── public/
+│       └── orchestrator-ui.js    # Frontend panel (~500 lines)
+```
+
+## Integration Points with Existing Code
+
+### StateStore (`src/state-store.ts`)
+```typescript
+// Add to AppState interface
+orchestrator?: OrchestratorPersistState;
+
+// Add methods
+getOrchestratorState(): OrchestratorPersistState;
+setOrchestratorState(state: Partial<OrchestratorPersistState>): void;
+```
+
+### SSE Events (`src/web/sse-events.ts`)
+```typescript
+// Add ~8 new events
+export const SseEvent = {
+  // ... existing
+  ORCHESTRATOR_STATE_CHANGED: 'orchestrator:stateChanged',
+  ORCHESTRATOR_PLAN_READY: 'orchestrator:planReady',
+  ORCHESTRATOR_PHASE_STARTED: 'orchestrator:phaseStarted',
+  ORCHESTRATOR_PHASE_COMPLETED: 'orchestrator:phaseCompleted',
+  ORCHESTRATOR_PHASE_FAILED: 'orchestrator:phaseFailed',
+  ORCHESTRATOR_VERIFICATION: 'orchestrator:verificationResult',
+  ORCHESTRATOR_COMPLETED: 'orchestrator:completed',
+  ORCHESTRATOR_ERROR: 'orchestrator:error',
+} as const;
+```
+
+### Frontend Constants (`src/web/public/constants.js`)
+```javascript
+// Mirror SSE events
+SSE_EVENTS.ORCHESTRATOR_STATE_CHANGED = 'orchestrator:stateChanged';
+// ... etc
+```
+
+### Route Registration (`src/web/routes/index.ts`)
+```typescript
+import { registerOrchestratorRoutes } from './orchestrator-routes.js';
+// Add to barrel export
+```
+
+### Server (`src/web/server.ts`)
+```typescript
+// Initialize OrchestratorLoop alongside RalphLoop
+const orchestratorLoop = new OrchestratorLoop(config);
+
+// Register routes
+registerOrchestratorRoutes(app, { ...ctx, orchestrator: orchestratorLoop });
+```
+
+### Port Interface (`src/web/ports/`)
+```typescript
+// New port
+export interface OrchestratorPort {
+  orchestrator: OrchestratorLoop;
+}
+```
+
+## Prompt Flow Through System
+
+The key insight is how prompts flow from Orchestrator → Session → Claude:
+
+```
+OrchestratorLoop decides to execute Phase 3, Task 2
+  │
+  ▼
+Converts OrchestratorTask to CreateTaskOptions:
+  {
+    prompt: "Implement the rate limiter middleware. Read src/middleware/auth.ts
+             for the pattern. Add to src/middleware/rate-limiter.ts. Must export
+             a Fastify plugin. When done: <promise>PHASE_3_TASK_2_DONE</promise>",
+    priority: 100,
+    dependencies: ["phase-3-task-1"],  // Must finish auth middleware first
+    completionPhrase: "PHASE_3_TASK_2_DONE",
+    timeoutMs: 600000  // 10 minutes
+  }
+  │
+  ▼
+TaskQueue.addTask(options)
+  │
+  ▼
+RalphLoop.tick() → assignTasks()  // OR OrchestratorLoop does its own assignment
+  │
+  ▼
+session.sendInput(task.prompt)
+  │
+  ▼
+writeViaMux() → tmux send-keys -l "prompt..." + Enter
+  │
+  ▼
+Claude CLI receives prompt, executes, outputs results
+  │
+  ▼
+RalphTracker.processData() → detects "PHASE_3_TASK_2_DONE"
+  │
+  ▼
+emit('completionDetected') → OrchestratorLoop.handleTaskCompleted()
+  │
+  ▼
+Check: all tasks in Phase 3 done? → If yes → verifyPhase(phase3)
+```
+
+## Team Agent Flow (When Enabled)
+
+```
+Phase has teamStrategy.type === 'team'
+  │
+  ▼
+OrchestratorLoop creates/reuses a session with:
+  env: { CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: '1' }
+  │
+  ▼
+Sends team orchestration prompt:
+  "You're the team lead for Phase 3: Core Implementation.
+
+   Your team should work on these tasks in parallel:
+   1. Rate limiter middleware (teammate 1)
+   2. Error handling middleware (teammate 2)
+   3. Validation layer (teammate 3)
+
+   Context files to read first: [...]
+   Each teammate should output their task's completion phrase when done.
+   When ALL tasks are complete, output: <promise>PHASE_3_COMPLETE</promise>"
+  │
+  ▼
+Claude Code team-lead spawns teammates
+  │
+  ▼
+TeamWatcher detects new team in ~/.claude/teams/
+  → Matches to session via leadSessionId
+  → Tracks teammate activity
+  │
+  ▼
+Teammates work in parallel (in-process threads)
+  │
+  ▼
+hook: teammate_idle → POST /api/hook-event
+  → OrchestratorLoop notes teammate finished
+  │
+  ▼
+hook: task_completed → POST /api/hook-event
+  → Or: RalphTracker detects PHASE_3_COMPLETE
+  → OrchestratorLoop → phase complete → verify
+```
+
+## Error Recovery Strategy
+
+```
+Task fails (timeout, error, session crash)
+  │
+  ├─ Task-level retry (up to 2 retries per task)
+  │   → Reset task to pending
+  │   → Re-queue with modified prompt: "Previous attempt failed: {error}. Try again..."
+  │
+  ├─ Phase-level retry (up to 3 retries per phase)
+  │   → Respawn session (fresh context)
+  │   → Re-execute entire phase with learnings from failure
+  │   → Modified prompt includes what went wrong
+  │
+  └─ Orchestration-level failure
+      → All retries exhausted
+      → state = FAILED
+      → Notify user with detailed failure report
+      → User can: modify plan → retry, skip phase → continue, or stop
+```
+
+## Interaction with Ralph Loop
+
+Ralph Loop and Orchestrator Loop are **mutually exclusive** on the same sessions:
+
+```
+if (orchestratorLoop.isRunning()) {
+  // Orchestrator controls task assignment
+  // Ralph Loop should not interfere
+  // Respawn Controller uses 'orchestrator' preset
+}
+
+if (ralphLoop.isRunning()) {
+  // Ralph controls task assignment
+  // Orchestrator should not start
+}
+```
+
+The Orchestrator can optionally USE the Ralph Loop internally for phase execution (delegate phase tasks to Ralph's queue), or manage task assignment directly. Decision: **manage directly** — gives more control over phase boundaries and verification timing.
+
+## Summary of What Touches What
+
+| Existing File | Change |
+|---|---|
+| `src/types/index.ts` | Export orchestrator types |
+| `src/state-store.ts` | Add orchestrator state persistence |
+| `src/web/sse-events.ts` | Add ~8 orchestrator events |
+| `src/web/routes/index.ts` | Register orchestrator routes |
+| `src/web/server.ts` | Initialize OrchestratorLoop |
+| `src/web/public/constants.js` | Mirror SSE events |
+| `src/web/public/app.js` | Add orchestrator event listeners, panel toggle |
+| `src/web/route-helpers.ts` | Add 'orchestrator' respawn preset |
+
+| New File | Purpose |
+|---|---|
+| `src/orchestrator-loop.ts` | Core state machine |
+| `src/orchestrator-planner.ts` | Plan generation + phasing |
+| `src/orchestrator-verifier.ts` | Phase verification |
+| `src/types/orchestrator.ts` | Type definitions |
+| `src/prompts/orchestrator.ts` | Prompt templates |
+| `src/web/routes/orchestrator-routes.ts` | API endpoints |
+| `src/web/public/orchestrator-ui.js` | Frontend panel |
+| `src/web/ports/orchestrator-port.ts` | Port interface |

docs/orchestrator-loop-plan.md

@@ -0,0 +1,633 @@
+# Orchestrator Loop — Detailed Implementation Plan (v2)
+
+> Internal research/planning document. Not for GitHub.
+
+## Vision
+
+The **Orchestrator Loop** is a new autonomous execution mode that transforms high-level user goals into phased, verified, team-coordinated implementations. Unlike Ralph Loop (flat task queue → idle sessions), the Orchestrator manages the full lifecycle: **plan → approve → execute → verify → adapt → complete**.
+
+```
+USER: "Add OAuth2 login with Google/GitHub, role-based access control, and API key management"
+
+ORCHESTRATOR:
+  Phase 1: Research & Setup      ✅ (3m)  — scaffold, deps, config
+  Phase 2: Auth Core             ✅ (8m)  — OAuth2 flow, session mgmt
+  Phase 3: Provider Integration  🔄 (12m) — Google + GitHub (parallel via team agents)
+  Phase 4: RBAC                  ⏳       — roles, permissions, middleware
+  Phase 5: API Keys              ⏳       — generation, validation, rate limits
+  Phase 6: Testing & Review      ⏳       — integration tests, security review
+
+  Progress: ━━━━━━━━━━━━━━━━━━━━ 40%   |   Agents: 3 active   |   Time: 23m
+```
+
+## Architecture
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│                    OrchestratorLoop                               │
+│                                                                  │
+│  ┌────────────────┐  ┌────────────────┐  ┌──────────────────┐  │
+│  │ Orchestrator   │  │ Orchestrator   │  │ Orchestrator     │  │
+│  │ Planner        │  │ Executor       │  │ Verifier         │  │
+│  │                │  │                │  │                  │  │
+│  │ PlanOrchestrator│ │ TaskQueue      │  │ AI review        │  │
+│  │ + phase grouper│  │ SessionManager │  │ Test commands    │  │
+│  │ + team strategy│  │ Team prompts   │  │ File checks      │  │
+│  └───────┬────────┘  └───────┬────────┘  └─────────┬────────┘  │
+│          │                   │                      │           │
+│          └───────────────────┼──────────────────────┘           │
+│                              │                                  │
+│                    ┌─────────▼─────────┐                       │
+│                    │  Existing Codeman  │                       │
+│                    │  Infrastructure    │                       │
+│                    │                    │                       │
+│                    │  SessionManager    │                       │
+│                    │  TaskQueue         │                       │
+│                    │  RespawnController │                       │
+│                    │  TeamWatcher       │                       │
+│                    │  PlanOrchestrator  │                       │
+│                    │  StateStore        │                       │
+│                    │  Hooks + SSE       │                       │
+│                    └────────────────────┘                       │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+## State Machine
+
+```
+                              ┌─────────┐
+                              │  IDLE   │
+                              └────┬────┘
+                                   │ start(goal)
+                                   ▼
+                              ┌─────────┐
+                     ┌────────│PLANNING │────────┐
+                     │ fail   └────┬────┘        │
+                     ▼             │ plan ready   │ user cancels
+                ┌────────┐        ▼              ▼
+                │ FAILED │   ┌─────────┐    ┌────────┐
+                └────────┘   │APPROVAL │    │ IDLE   │
+                     ▲       └────┬────┘    └────────┘
+                     │            │ approve
+                     │            ▼
+                     │       ┌──────────┐
+                     │  ┌───►│EXECUTING │◄────────────────────┐
+                     │  │    └────┬─────┘                     │
+                     │  │         │ all tasks in phase done    │
+                     │  │         ▼                            │
+                     │  │    ┌──────────┐                     │
+                     │  │    │VERIFYING │                     │
+                     │  │    └────┬─────┘                     │
+                     │  │    pass │    │ fail                  │
+                     │  │         ▼    ▼                       │
+                     │  │   more  ┌──────────┐                │
+                     │  │  phases?│REPLANNING│── retry ────────┘
+                     │  │    │    └────┬─────┘
+                     │  │    │         │ max retries
+                     │  │    │         ▼
+                     │  │    │    ┌────────┐
+                     │  └────┘    │ FAILED │
+                     │  next      └────────┘
+                     │  phase
+                     │    │
+                     │    ▼
+                     │ ┌───────────┐
+                     └─│ COMPLETED │
+                       └───────────┘
+```
+
+**States:** `idle` | `planning` | `approval` | `executing` | `verifying` | `replanning` | `completed` | `failed` | `paused`
+
+Transitions are event-driven. The state machine is the single source of truth — all methods check `this.state` before acting.
+
+## Type Definitions
+
+### `src/types/orchestrator.ts`
+
+```typescript
+// ═══════════════════════════════════════════════════════════════
+// State Machine
+// ═══════════════════════════════════════════════════════════════
+
+export type OrchestratorState =
+  | 'idle'
+  | 'planning'
+  | 'approval'
+  | 'executing'
+  | 'verifying'
+  | 'replanning'
+  | 'completed'
+  | 'failed'
+  | 'paused';
+
+// ═══════════════════════════════════════════════════════════════
+// Plan Structure
+// ═══════════════════════════════════════════════════════════════
+
+export interface OrchestratorPlan {
+  id: string;
+  goal: string;
+  createdAt: number;
+  phases: OrchestratorPhase[];
+  metadata: {
+    totalTasks: number;
+    estimatedComplexity: 'low' | 'medium' | 'high';
+    modelUsed: string;
+    planDurationMs: number;
+  };
+}
+
+export interface OrchestratorPhase {
+  id: string;                      // "phase-1", "phase-2"
+  name: string;                    // Human-readable name
+  description: string;
+  order: number;
+  status: PhaseStatus;
+  tasks: OrchestratorTask[];
+  verificationCriteria: string[];
+  testCommands: string[];
+  maxAttempts: number;             // Default: 3
+  attempts: number;                // Current attempt count
+  startedAt: number | null;
+  completedAt: number | null;
+  durationMs: number | null;
+  teamStrategy: TeamStrategy;
+}
+
+export type PhaseStatus =
+  | 'pending'
+  | 'executing'
+  | 'verifying'
+  | 'passed'
+  | 'failed'
+  | 'skipped';
+
+export interface OrchestratorTask {
+  id: string;                      // "phase-1-task-1"
+  phaseId: string;
+  prompt: string;                  // Single-line prompt for Claude
+  status: 'pending' | 'running' | 'completed' | 'failed';
+  assignedSessionId: string | null;
+  queueTaskId: string | null;     // Links to TaskQueue task
+  parallel: boolean;               // Can run in parallel with sibling tasks
+  completionPhrase: string;        // Unique phrase for completion detection
+  timeoutMs: number;
+  startedAt: number | null;
+  completedAt: number | null;
+  error: string | null;
+  retries: number;
+}
+
+// ═══════════════════════════════════════════════════════════════
+// Team Strategy
+// ═══════════════════════════════════════════════════════════════
+
+export type TeamStrategy =
+  | { type: 'single' }                          // One session handles all
+  | { type: 'parallel'; maxSessions: number }   // Multiple sessions
+  | { type: 'team'; config: TeamSetup }          // Agent teams
+
+export interface TeamSetup {
+  leadPrompt: string;
+  suggestedTeammates: string[];    // Role descriptions
+  maxTeammates: number;
+}
+
+// ═══════════════════════════════════════════════════════════════
+// Verification
+// ═══════════════════════════════════════════════════════════════
+
+export interface VerificationResult {
+  passed: boolean;
+  checks: VerificationCheck[];
+  summary: string;
+  suggestions: string[];           // Recovery hints for replanning
+}
+
+export interface VerificationCheck {
+  type: 'test_command' | 'ai_review' | 'file_check';
+  description: string;
+  passed: boolean;
+  output?: string;
+}
+
+// ═══════════════════════════════════════════════════════════════
+// Configuration
+// ═══════════════════════════════════════════════════════════════
+
+export interface OrchestratorConfig {
+  plannerModel: string;            // Default: 'opus'
+  researchEnabled: boolean;        // Default: true
+  autoApprove: boolean;            // Default: false
+  maxPhaseRetries: number;         // Default: 3
+  phaseTimeoutMs: number;          // Default: 1800000 (30min)
+  enableTeamAgents: boolean;       // Default: true
+  maxParallelSessions: number;     // Default: 3
+  verificationMode: 'strict' | 'moderate' | 'lenient';
+  compactBetweenPhases: boolean;   // Default: true
+}
+
+// ═══════════════════════════════════════════════════════════════
+// Persistence (saved to ~/.codeman/state.json)
+// ═══════════════════════════════════════════════════════════════
+
+export interface OrchestratorPersistState {
+  state: OrchestratorState;
+  plan: OrchestratorPlan | null;
+  currentPhaseIndex: number;
+  startedAt: number | null;
+  completedAt: number | null;
+  config: OrchestratorConfig;
+  stats: OrchestratorStats;
+}
+
+export interface OrchestratorStats {
+  phasesCompleted: number;
+  phasesFailed: number;
+  totalTasksCompleted: number;
+  totalTasksFailed: number;
+  totalDurationMs: number;
+  replanCount: number;
+}
+```
+
+## New Files (Implementation Order)
+
+### Step 1: `src/types/orchestrator.ts` — Type definitions
+All interfaces above. No dependencies. ~120 lines.
+
+### Step 2: `src/orchestrator-planner.ts` — Plan generation + phase grouping
+~300 lines. Wraps existing PlanOrchestrator.
+
+```typescript
+/**
+ * @fileoverview Orchestrator plan generation — converts goals into phased plans.
+ *
+ * Uses PlanOrchestrator for AI plan generation, then groups PlanItems into
+ * sequential phases with team strategies and verification criteria.
+ *
+ * @module orchestrator-planner
+ */
+
+export class OrchestratorPlanner {
+  constructor(mux: TerminalMultiplexer, workingDir: string, config: OrchestratorConfig);
+
+  /** Generate plan from goal. Uses PlanOrchestrator internally. */
+  async generatePlan(goal: string, onProgress?: ProgressCallback): Promise<OrchestratorPlan>;
+
+  /** Cancel in-progress plan generation. */
+  async cancel(): Promise<void>;
+
+  // Internal
+  private groupIntoPhases(items: PlanItem[], goal: string): OrchestratorPhase[];
+  private assignTeamStrategies(phases: OrchestratorPhase[]): void;
+  private generateCompletionPhrases(plan: OrchestratorPlan): void;
+}
+```
+
+**Phase grouping algorithm:**
+1. Topological sort by `PlanItem.dependencies`
+2. Group into dependency layers (Kahn's algorithm)
+3. Within each layer, sub-group by `tddPhase` (setup → test → impl → verify → review)
+4. Merge adjacent small phases (< 2 tasks) if they share the same tddPhase
+5. Assign team strategies:
+   - 1-2 tasks → `{ type: 'single' }`
+   - 3+ independent tasks → `{ type: 'parallel', maxSessions: Math.min(taskCount, config.maxParallelSessions) }`
+   - 4+ tasks with high complexity → `{ type: 'team', config: { ... } }`
+6. Generate unique completion phrases per task: `ORCH_P{phaseOrder}_T{taskIndex}`
+
+### Step 3: `src/orchestrator-verifier.ts` — Phase verification
+~200 lines.
+
+```typescript
+/**
+ * @fileoverview Orchestrator phase verification.
+ *
+ * Runs verification checks after each phase completes:
+ * test commands, AI review, and file existence checks.
+ *
+ * @module orchestrator-verifier
+ */
+
+export class OrchestratorVerifier {
+  constructor(config: OrchestratorConfig);
+
+  /** Run all verification checks for a completed phase. */
+  async verifyPhase(
+    phase: OrchestratorPhase,
+    session: Session,
+    mode: 'strict' | 'moderate' | 'lenient'
+  ): Promise<VerificationResult>;
+
+  // Verification strategies
+  private async runTestCommands(commands: string[], session: Session): Promise<VerificationCheck[]>;
+  private async aiReview(phase: OrchestratorPhase, session: Session): Promise<VerificationCheck>;
+}
+```
+
+**Verification modes:**
+- `strict`: ALL test commands must pass AND AI review must approve
+- `moderate`: Test commands must pass, AI review is advisory
+- `lenient`: At least one test command passes, AI review skipped
+
+**AI review prompt (sent as a task to the session):**
+```
+Review Phase "{phase.name}" completion. Check:
+1. Expected functionality works
+2. No obvious regressions
+3. Code quality is acceptable
+
+Criteria: {phase.verificationCriteria.join('\n')}
+
+If ALL criteria are met, respond: ORCH_VERIFY_PASS
+If ANY criteria fail, respond: ORCH_VERIFY_FAIL and explain what failed.
+```
+
+### Step 4: `src/orchestrator-loop.ts` — Core state machine
+~500 lines. Main orchestrator engine.
+
+```typescript
+/**
+ * @fileoverview Orchestrator Loop — phased plan execution with team agents.
+ *
+ * State machine that generates plans from user goals, executes them
+ * phase-by-phase with verification gates, and adapts on failure.
+ *
+ * @module orchestrator-loop
+ */
+
+export interface OrchestratorLoopEvents {
+  stateChanged: (state: OrchestratorState, prevState: OrchestratorState) => void;
+  planReady: (plan: OrchestratorPlan) => void;
+  phaseStarted: (phase: OrchestratorPhase) => void;
+  phaseCompleted: (phase: OrchestratorPhase) => void;
+  phaseFailed: (phase: OrchestratorPhase, reason: string) => void;
+  taskAssigned: (task: OrchestratorTask, sessionId: string) => void;
+  taskCompleted: (task: OrchestratorTask) => void;
+  taskFailed: (task: OrchestratorTask, error: string) => void;
+  verificationResult: (phase: OrchestratorPhase, result: VerificationResult) => void;
+  completed: (stats: OrchestratorStats) => void;
+  error: (error: Error) => void;
+}
+
+export class OrchestratorLoop extends EventEmitter {
+  private state: OrchestratorState = 'idle';
+  private plan: OrchestratorPlan | null = null;
+  private currentPhaseIndex = 0;
+  private config: OrchestratorConfig;
+  private planner: OrchestratorPlanner;
+  private verifier: OrchestratorVerifier;
+  private sessionManager: SessionManager;
+  private taskQueue: TaskQueue;
+  private store: StateStore;
+  private stats: OrchestratorStats;
+  private cleanup: CleanupManager;
+  private pausedState: OrchestratorState | null = null; // State before pause
+
+  // ── Lifecycle ──────────────────────────────────────────────
+
+  constructor(mux: TerminalMultiplexer, workingDir: string, config?: Partial<OrchestratorConfig>);
+
+  /** Start orchestration with a goal. Transitions: idle → planning */
+  async start(goal: string): Promise<void>;
+
+  /** Approve the generated plan. Transitions: approval → executing */
+  async approve(): Promise<void>;
+
+  /** Reject plan with feedback. Transitions: approval → planning (regenerate) */
+  async reject(feedback: string): Promise<void>;
+
+  /** Pause execution. Saves current state. */
+  pause(): void;
+
+  /** Resume from pause. */
+  resume(): void;
+
+  /** Stop everything and clean up. → idle */
+  async stop(): Promise<void>;
+
+  /** Skip current phase. → executing (next phase) or completed */
+  async skipPhase(phaseId: string): Promise<void>;
+
+  /** Retry a failed phase. → executing */
+  async retryPhase(phaseId: string): Promise<void>;
+
+  // ── Getters ────────────────────────────────────────────────
+
+  getState(): OrchestratorState;
+  getPlan(): OrchestratorPlan | null;
+  getCurrentPhase(): OrchestratorPhase | null;
+  getStats(): OrchestratorStats;
+  getStatus(): OrchestratorPersistState;
+
+  // ── Internal: Phase Execution ──────────────────────────────
+
+  private async executeCurrentPhase(): Promise<void>;
+  private async executePhase(phase: OrchestratorPhase): Promise<void>;
+  private async assignPhaseTasks(phase: OrchestratorPhase): Promise<void>;
+  private handleTaskCompleted(taskId: string): void;
+  private handleTaskFailed(taskId: string, error: string): void;
+  private async onPhaseTasksComplete(phase: OrchestratorPhase): Promise<void>;
+
+  // ── Internal: Verification ─────────────────────────────────
+
+  private async verifyCurrentPhase(): Promise<void>;
+  private async handleVerificationResult(phase: OrchestratorPhase, result: VerificationResult): Promise<void>;
+
+  // ── Internal: Replanning ───────────────────────────────────
+
+  private async replanPhase(phase: OrchestratorPhase, failures: string[]): Promise<void>;
+
+  // ── Internal: State Machine ────────────────────────────────
+
+  private setState(newState: OrchestratorState): void;
+  private advanceToNextPhase(): Promise<void>;
+  private persist(): void;
+  private restore(): void;
+}
+```
+
+**Key execution flow in `executePhase()`:**
+1. Mark phase as `executing`, emit `phaseStarted`
+2. For each task in phase:
+   - Create a `CreateTaskOptions` from `OrchestratorTask`
+   - Add to `TaskQueue` with proper dependencies + completion phrase
+   - Store the TaskQueue task ID in `OrchestratorTask.queueTaskId`
+3. Poll task completion (listen to TaskQueue events)
+4. When all tasks complete → call `onPhaseTasksComplete()`
+5. `onPhaseTasksComplete()` triggers verification
+
+**How tasks get assigned to sessions:**
+The OrchestratorLoop does NOT manage session assignment directly. It adds tasks to the existing TaskQueue and starts a mini poll loop that assigns pending tasks to idle sessions — the same pattern as RalphLoop's `assignTasks()`. This reuses existing session management.
+
+**Team agent flow:**
+For phases with `teamStrategy.type === 'team'`:
+- Start a single session with `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`
+- Instead of adding individual tasks to TaskQueue, send ONE comprehensive prompt to the lead
+- The prompt instructs the lead to create teammates and delegate
+- Monitor via TeamWatcher for team task completion + hook events
+- Phase completion is detected via the lead's completion phrase
+
+### Step 5: `src/web/routes/orchestrator-routes.ts` — API endpoints
+~300 lines.
+
+```
+POST   /api/orchestrator/start          — { goal, config? } → start planning
+POST   /api/orchestrator/approve        — approve generated plan
+POST   /api/orchestrator/reject         — { feedback } → reject + replan
+POST   /api/orchestrator/pause          — pause execution
+POST   /api/orchestrator/resume         — resume execution
+POST   /api/orchestrator/stop           — stop orchestration
+GET    /api/orchestrator/status         — full state + plan + stats
+GET    /api/orchestrator/plan           — plan details only
+POST   /api/orchestrator/phase/:id/skip — skip a phase
+POST   /api/orchestrator/phase/:id/retry — retry a failed phase
+```
+
+Port dependency: `SessionPort & EventPort & RespawnPort & ConfigPort & InfraPort`
+
+The route module receives the OrchestratorLoop instance via the InfraPort (added to `createRouteContext()`).
+
+### Step 6: SSE Events — `src/web/sse-events.ts` additions
+
+```typescript
+// ─── Orchestrator ────────────────────────────────────────────────────────────
+
+/** Orchestrator state machine transitioned. */
+export const OrchestratorStateChanged = 'orchestrator:stateChanged' as const;
+/** Orchestrator plan generated and ready for approval. */
+export const OrchestratorPlanReady = 'orchestrator:planReady' as const;
+/** Orchestrator phase started executing. */
+export const OrchestratorPhaseStarted = 'orchestrator:phaseStarted' as const;
+/** Orchestrator phase completed successfully. */
+export const OrchestratorPhaseCompleted = 'orchestrator:phaseCompleted' as const;
+/** Orchestrator phase failed. */
+export const OrchestratorPhaseFailed = 'orchestrator:phaseFailed' as const;
+/** Orchestrator verification result for a phase. */
+export const OrchestratorVerification = 'orchestrator:verification' as const;
+/** Orchestrator task assigned to session. */
+export const OrchestratorTaskAssigned = 'orchestrator:taskAssigned' as const;
+/** Orchestrator task completed. */
+export const OrchestratorTaskCompleted = 'orchestrator:taskCompleted' as const;
+/** Orchestrator task failed. */
+export const OrchestratorTaskFailed = 'orchestrator:taskFailed' as const;
+/** All phases completed successfully. */
+export const OrchestratorCompleted = 'orchestrator:completed' as const;
+/** Orchestrator error. */
+export const OrchestratorError = 'orchestrator:error' as const;
+```
+
+11 new events. Add to `SseEvent` namespace object + mirror in `constants.js`.
+
+### Step 7: State persistence — `src/state-store.ts` additions
+
+Add to `AppState`:
+```typescript
+orchestrator?: OrchestratorPersistState;
+```
+
+Add methods:
+```typescript
+getOrchestratorState(): OrchestratorPersistState | null;
+setOrchestratorState(state: Partial<OrchestratorPersistState>): void;
+clearOrchestratorState(): void;
+```
+
+### Step 8: Server integration — `src/web/server.ts` modifications
+
+1. Import `OrchestratorLoop` and `registerOrchestratorRoutes`
+2. Add `private orchestratorLoop: OrchestratorLoop` field
+3. Initialize in constructor (lazy — created on first start, not at boot)
+4. Add to `createRouteContext()` InfraPort: `orchestratorLoop: this.orchestratorLoop`
+5. Wire up OrchestratorLoop events → SSE broadcasts
+6. Register routes: `registerOrchestratorRoutes(this.app, ctx)`
+7. Clean up in `stop()`
+
+### Step 9: `src/web/public/orchestrator-ui.js` — Frontend panel
+~500 lines. New frontend module.
+
+**Load order**: After `panels-ui.js` (11), before `ralph-wizard.js` (13). So load order = 11.5.
+
+**UI elements:**
+- Goal input form (text area + config toggles)
+- Plan approval view (phase list, task details, approve/reject buttons)
+- Execution dashboard (progress bar, phase cards, task status indicators)
+- Agent activity panel (session count, team status)
+- Controls (pause, resume, stop, skip phase, retry phase)
+
+**SSE listeners:**
+- All 11 orchestrator events → update UI state
+- Reuses existing session/respawn/team event handlers for agent monitoring
+
+### Step 10: `src/prompts/orchestrator.ts` — Prompt templates
+~200 lines.
+
+Templates for:
+- Phase execution prompt (tells Claude what to do in this phase)
+- Team lead delegation prompt (instructs lead to create and coordinate teammates)
+- Verification prompt (asks Claude to verify phase output)
+- Replan prompt (gives failure context, asks for recovery steps)
+
+### Step 11: Constants, schemas, route barrel updates
+
+- `src/web/public/constants.js` — Add 11 SSE event mirrors
+- `src/web/schemas.ts` — Add Zod schemas for orchestrator API input validation
+- `src/web/routes/index.ts` — Export `registerOrchestratorRoutes`
+- `src/web/ports/infra-port.ts` — Add `orchestratorLoop` to InfraPort
+- `src/types/index.ts` — Export orchestrator types
+
+## Existing File Modifications Summary
+
+| File | Change | Lines |
+|------|--------|-------|
+| `src/types/index.ts` | Add orchestrator barrel export | +1 |
+| `src/web/sse-events.ts` | Add 11 orchestrator events + SseEvent entries | +30 |
+| `src/web/public/constants.js` | Mirror 11 SSE events | +15 |
+| `src/web/routes/index.ts` | Export registerOrchestratorRoutes | +1 |
+| `src/web/ports/infra-port.ts` | Add orchestratorLoop to InfraPort | +3 |
+| `src/web/server.ts` | Initialize OrchestratorLoop, wire events, register routes | +40 |
+| `src/web/schemas.ts` | Add orchestrator Zod schemas | +20 |
+| `src/state-store.ts` | Add orchestrator state persistence | +20 |
+| `src/web/public/app.js` | Add orchestrator SSE listeners + panel toggle | +30 |
+| `src/web/public/index.html` | Add orchestrator-ui.js script tag | +1 |
+
+**Total new code**: ~2,300 lines across 6 new files
+**Total modifications**: ~160 lines across 10 existing files
+
+## Implementation Execution Order
+
+This is the actual build order — each step is a commit checkpoint:
+
+1. **Types** — `src/types/orchestrator.ts` + barrel export. Zero risk, pure types.
+2. **SSE events** — Add all 11 events to both `sse-events.ts` and `constants.js`. Wire in SseEvent namespace.
+3. **State persistence** — Add orchestrator state to StateStore. Small, isolated change.
+4. **Schemas** — Add Zod validation schemas for API input.
+5. **Planner** — `src/orchestrator-planner.ts`. Can test in isolation.
+6. **Verifier** — `src/orchestrator-verifier.ts`. Can test in isolation.
+7. **Core loop** — `src/orchestrator-loop.ts`. The big one. Depends on planner + verifier.
+8. **Prompts** — `src/prompts/orchestrator.ts`. Templates used by core loop.
+9. **Port + routes** — `src/web/ports/infra-port.ts` update + `src/web/routes/orchestrator-routes.ts`.
+10. **Server integration** — Wire OrchestratorLoop into WebServer. Routes become live.
+11. **Frontend** — `src/web/public/orchestrator-ui.js` + app.js listeners + index.html script tag.
+12. **Tests** — `test/orchestrator-*.test.ts`.
+13. **Typecheck + lint** — Fix all issues, ensure CI passes.
+
+## Edge Cases & Error Handling
+
+- **Session limit reached**: Queue tasks and wait for sessions to free up (existing SessionManager handles this)
+- **All sessions crash during phase**: Mark phase as failed, attempt replan
+- **Verification flaky**: `moderate` mode allows test retries; `lenient` skips AI review
+- **Plan too large**: Cap at 10 phases, 50 total tasks. Warn user.
+- **Context overflow**: Auto-compact between phases. Respawn if needed (orchestrator state is external).
+- **User pauses mid-phase**: Pause task assignment, don't cancel running tasks. Resume picks up where it left off.
+- **Network/API errors during planning**: Retry plan generation up to 2 times, then fail with clear message.
+- **Orchestrator vs Ralph conflict**: Mutually exclusive. Starting orchestrator stops Ralph if running. Starting Ralph stops orchestrator.
+
+## Testing Strategy
+
+- **Unit tests**: `test/orchestrator-planner.test.ts` — phase grouping algorithm, team strategy assignment
+- **Unit tests**: `test/orchestrator-verifier.test.ts` — verification logic with mocked sessions
+- **Integration tests**: `test/orchestrator-loop.test.ts` — state machine transitions, task lifecycle
+- **Route tests**: `test/routes/orchestrator-routes.test.ts` — API validation, status responses
+
+All tests use `MockSession` pattern from existing test infrastructure. No real tmux needed.

docs/orchestrator-loop-research.md

@@ -0,0 +1,157 @@
+# Orchestrator Loop — Research Findings
+
+> Research doc for the new "Orchestrator Loop" feature. Not for GitHub.
+
+## What We're Building
+
+A new autonomous loop variant — **Orchestrator Loop** — that takes high-level user tasks, decomposes them into a detailed plan using team agents, and executes the plan step-by-step with quality gates. Unlike Ralph Loop (which executes a flat task queue), the Orchestrator coordinates **planning, delegation, and verification** as a continuous cycle.
+
+**Core idea**: User inputs a goal → Orchestrator creates a detailed plan → spins up team agents for parallel execution → validates each step → adapts the plan based on results → delivers polished output.
+
+## Existing Infrastructure Analysis
+
+### What We Can Reuse
+
+#### 1. Ralph Loop (`src/ralph-loop.ts`)
+- **Pattern**: Poll loop with `start() → tick() → stop()` lifecycle
+- **Reusable**: Event-driven task assignment, session completion handling, timeout management
+- **Limitation**: Flat task queue — no concept of phases, dependencies between task groups, or adaptive replanning
+- **Key insight**: `assignTaskToSession()` uses `session.sendInput(task.prompt)` — simple prompt injection into PTY
+
+#### 2. Task Queue (`src/task-queue.ts`) + Task (`src/task.ts`)
+- **Already has**: Priority ordering, dependency tracking between tasks, completion phrase detection
+- **Limitation**: No task *groups* or *phases*. Dependencies are task-to-task, not phase-to-phase
+- **Key insight**: Tasks support `completionPhrase` — a string the task watches for in output. This is how Ralph knows a task is done
+
+#### 3. Plan Orchestrator (`src/plan-orchestrator.ts`)
+- **Already has**: 2-agent plan generation (Research Agent → Planner Agent), TDD-aware plan items with P0/P1/P2 priorities
+- **Output**: `PlanItem[]` with dependencies, verification criteria, TDD phases, complexity ratings
+- **Limitation**: Plan generation only — no execution. Plans are generated then sit in state/UI for human review
+- **Key insight**: Uses `Session` directly to run Claude subagent instances for research and planning. Returns structured JSON
+
+#### 4. Team Agents (`src/team-watcher.ts`, `~/.claude/teams/`)
+- **Already has**: Team creation, member tracking, filesystem inbox messaging, task management via `~/.claude/tasks/{team-name}/`
+- **Limitation**: Codeman can only *observe* teams (TeamWatcher is read-only polling), not *create* or *orchestrate* them
+- **Key insight**: Teams are a Claude Code feature. Codeman monitors them but doesn't control them. We can't programmatically create teammates — Claude Code does that when you use `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`
+
+#### 5. Respawn Controller (`src/respawn-controller.ts`)
+- **Already has**: Preset-based automation (ralph-todo, overnight-autonomous), circuit breaker, health scoring
+- **Key insight**: The `ralph-todo` preset (8s idle, 480min max) is designed for autonomous task execution. We'd need a new preset or make Orchestrator Loop set its own timing
+
+#### 6. Session Auto-Ops (`src/session-auto-ops.ts`)
+- **Already has**: Auto-compact at token thresholds, auto-clear for context management
+- **Key insight**: Critical for long Orchestrator runs — prevents context overflow during multi-step execution
+
+#### 7. Hooks (`src/hooks-config.ts`)
+- **Already has**: `idle_prompt`, `stop`, `teammate_idle`, `task_completed` hook events
+- **Key insight**: Hooks fire POST to `/api/hook-event` — this is how Codeman knows when Claude is idle, stopped, or completed a task. The Orchestrator Loop can listen to these same events
+
+### What We Need to Build New
+
+1. **Plan → Task decomposition**: Convert PlanOrchestrator output (PlanItem[]) into executable task groups with phase ordering
+2. **Multi-phase execution engine**: Execute plan phases sequentially, tasks within phases in parallel
+3. **Verification gates**: After each phase, run verification (test commands, AI review) before proceeding
+4. **Adaptive replanning**: When a task fails or verification fails, generate a recovery plan
+5. **Team agent orchestration**: Leverage Claude Code's agent teams for parallel execution within phases
+6. **Progress tracking & UI**: Real-time dashboard showing plan progress, phase status, agent activity
+
+## How Teams Actually Work (Important Constraint)
+
+After deep research, here's the reality of agent teams:
+
+```
+User starts session with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
+  → Claude Code creates a team-lead
+  → Team-lead spawns teammates (in-process threads)
+  → Teammates appear as subagents (detected by SubagentWatcher)
+  → Communication via ~/.claude/teams/{name}/inboxes/{member}.json
+  → Tasks tracked in ~/.claude/tasks/{team-name}/{N}.json
+```
+
+**Codeman cannot programmatically create team members.** This is a Claude Code internal feature. However, Codeman CAN:
+- Start a session that has teams enabled
+- Send a prompt to the lead that instructs it to use agent teams
+- Monitor team activity via TeamWatcher
+- React to teammate_idle and task_completed hook events
+- Read team task status from the filesystem
+
+**This means**: The Orchestrator Loop orchestrates at the *session prompt* level, not the *team member* level. We tell the lead what to do, and the lead decides how to use its team.
+
+## Architecture Decision: Prompt-Level Orchestration
+
+Given the team constraint, the Orchestrator Loop works by:
+
+1. **Planning phase**: Use PlanOrchestrator to generate a detailed plan from user input
+2. **Execution phase**: Feed plan steps as prompts to sessions, one phase at a time
+3. **Verification phase**: After each phase, run verification prompts and check results
+4. **Adaptation phase**: If verification fails, generate recovery prompts
+
+The "team agents" aspect works by:
+- Starting sessions with `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`
+- Crafting prompts that *instruct the lead to delegate* to teammates
+- Monitoring team activity to track parallel progress
+- The lead agent is smart enough to decompose work across its team
+
+## Key Technical Findings
+
+### Session Input Mechanics
+```typescript
+// From session.ts - how we send prompts
+await session.sendInput(task.prompt);  // Uses writeViaMux() internally
+// writeViaMux() does: tmux send-keys -l "prompt text" + tmux send-keys Enter
+// CRITICAL: Single-line only! Multi-line breaks Ink rendering
+```
+
+### Completion Detection Chain
+```
+PTY output → RalphTracker.processData() → completion phrase fuzzy match
+  → CompletionConfidence scoring (multi-signal: promise tag + todos + exit signal)
+  → If confident → emit 'completionDetected'
+  → RalphLoop listens → marks task complete → assigns next
+```
+
+### How Plan Items Map to Tasks
+```typescript
+// PlanItem has:
+interface PlanItem {
+  id: string;            // "P0-001"
+  content: string;       // "Implement error handling for API endpoints"
+  priority: 'P0' | 'P1' | 'P2';
+  dependencies: string[]; // ["P0-000"] — other PlanItem IDs
+  verificationCriteria: string;
+  testCommand: string;
+  tddPhase: 'setup' | 'test' | 'impl' | 'verify' | 'review';
+  complexity: 'low' | 'medium' | 'high';
+}
+
+// Task has:
+interface CreateTaskOptions {
+  prompt: string;
+  priority: number;
+  dependencies: string[];  // Task IDs
+  completionPhrase: string;
+  timeoutMs: number;
+}
+
+// Natural mapping: PlanItem.content → Task.prompt
+// PlanItem.dependencies → Task.dependencies
+// PlanItem.priority → Task.priority (P0=100, P1=50, P2=10)
+// PlanItem.verificationCriteria → verification task prompt
+```
+
+### Context Management for Long Runs
+- Auto-compact at ~110k tokens (configurable)
+- Auto-clear at ~140k tokens (configurable)
+- Respawn cycling: kill + restart session to reset context entirely
+- For Orchestrator: we want compact between phases, respawn between major milestones
+
+## Risk Assessment
+
+| Risk | Severity | Mitigation |
+|------|----------|------------|
+| Context overflow during complex phases | High | Auto-compact between tasks, respawn between phases |
+| Team agents not predictable | Medium | Orchestrate at session level, let Claude decide team delegation |
+| Plan too ambitious → infinite loop | High | Phase budgets (max attempts per phase), circuit breaker |
+| Verification too strict → blocks progress | Medium | Configurable strictness, human override via UI |
+| Single-line prompt limit | Medium | Use CLAUDE.md file for complex instructions, prompt references file |
+| Long planning phase delays execution | Low | Show plan for approval before execution |

docs/performance-analysis.md

@@ -0,0 +1,423 @@
+# Performance Analysis & Optimization Opportunities
+
+**Date**: 2026-03-07
+**Scope**: Full-stack performance analysis — backend PTY handling, SSE broadcasting, frontend terminal rendering, local echo overlay, DOM updates, config/scaling limits.
+**Constraint**: All recommendations preserve existing functionality including local echo, backpressure, anti-flicker pipeline, and mobile support.
+
+---
+
+## Executive Summary
+
+The codebase is already well-optimized in critical paths. The multi-layer backpressure system, adaptive terminal batching, DEC 2026 sync markers, and incremental state serialization are strong. The main opportunities are in **reducing unnecessary work** (SSE filtering, DOM rebuilds, lazy terminal init) rather than algorithmic changes.
+
+**Top 5 high-impact opportunities:**
+
+| # | Optimization | Impact | Risk | Effort |
+|---|-------------|--------|------|--------|
+| 1 | Session-scoped SSE subscriptions | Bandwidth -60-80%, CPU -40% | Medium | Medium |
+| 2 | Lazy xterm.js for minimized subagent windows | Memory -3.5MB at 50 agents | Low | Low |
+| 3 | Targeted badge update (skip full tab rebuild) | Eliminates O(n) reflow on badge change | Low | Low |
+| 4 | Conditional SSE padding (tunnel-only, terminal-only) | Bandwidth -70% when tunneled | Low | Low |
+| 5 | Canvas renderer on mobile | GPU pressure reduction, battery savings | Low | Low |
+
+---
+
+## 1. SSE Broadcasting
+
+### Current State
+- **92 event types** broadcast to all connected clients (max 100)
+- Single `JSON.stringify()` per event, shared across all clients (efficient)
+- **No per-client filtering** — every client receives every event regardless of which session they're viewing
+- 8KB padding appended to **every** event when tunnel is active (forces Cloudflare proxy flush)
+- Backpressure: clients marked as backpressured if `reply.raw.write()` returns false; recovery via `session:needsRefresh`
+
+### Bottlenecks
+
+**B1: No session-scoped SSE subscriptions** (`server.ts:1986`)
+- Client viewing session A still receives all events for sessions B through T
+- With 20 active sessions, ~95% of terminal events are irrelevant to any given client
+- Cost: wasted bandwidth, CPU for JSON parsing, and event handler dispatch on client
+
+**B2: Unconditional 8KB padding** (`server.ts:1977`)
+- Every event gets 8KB comment padding when tunnel is active
+- A `task:updated` event (~200 bytes payload) becomes ~8.2KB
+- High-frequency events like `session:terminal` need the padding; low-frequency events like `session:created` don't
+
+### Recommendations
+
+**R1: Session-scoped SSE subscriptions** (High impact)
+- Add `?sessions=id1,id2` query param to `/api/events` SSE endpoint
+- Server filters events by session ID before broadcasting
+- Client subscribes to active session + "global" events (session lifecycle, system)
+- Re-subscribes on tab switch (or subscribe to all with client-side filter as fallback)
+- **Savings**: ~80% bandwidth reduction for single-session viewers; ~60% for multi-session dashboards
+
+**R2: Tiered SSE padding** (Medium impact)
+- Only pad `session:terminal` events and SSE heartbeats (the two that need proxy flush)
+- Skip padding for low-frequency structural events (`session:created`, `task:updated`, etc.)
+- **Savings**: ~70% padding overhead reduction; terminal events already large enough to flush
+
+---
+
+## 2. Terminal Rendering
+
+### Current State (Well-Optimized)
+- **6-layer anti-flicker pipeline**: Server batching (adaptive 16-50ms) → DEC 2026 sync wrap → single JSON serialize → client rAF batching → sync segment parser → chunked buffer loading (32KB/frame)
+- **64KB/frame write budget** with DEC 2026 sync-segment awareness (prevents 141KB single-frame freezes)
+- **3-layer backpressure**: SSE cap (128KB queued → drop + refresh), frame budget (64KB/frame), chunked restore (32KB/frame)
+- WebGL renderer enabled by default with canvas fallback on context loss
+- Typical latency: 16-32ms; worst case: ~115ms (50ms server batch + 50ms sync wait + 16ms rAF)
+
+### Bottlenecks
+
+**B3: WebGL on mobile** (`app.js:627-637`)
+- Mobile GPUs are weaker; WebGL context loss more likely on low-end devices
+- Canvas renderer is sufficient for mobile (typically 1 session, smaller viewport)
+
+**B4: Static scrollback for all sessions** (`app.js:572`)
+- Default 5000 lines scrollback for all sessions regardless of activity level
+- Heavy output sessions (build logs, test runners) accumulate large scroll buffers
+
+**B5: No addon lazy loading**
+- FitAddon, Unicode11Addon, and WebGLAddon all loaded at terminal init
+- Unicode11Addon only needed for CJK content; WebGLAddon is large
+
+### Recommendations
+
+**R3: Force canvas renderer on mobile** (Low risk)
+- Detect `MobileDetection.isMobile()` and skip WebGL addon loading
+- Reduces GPU memory pressure, prevents context loss crashes
+- Mobile typically has 1-2 sessions — canvas performance is more than adequate
+
+**R4: Dynamic scrollback based on session activity** (Low risk)
+- Active sessions (working state): 5000 lines (current default)
+- Inactive/idle sessions: reduce to 2000 lines
+- Restore on session select (fetch from server buffer)
+- **Savings**: ~60% scrollback memory for idle sessions
+
+**R5: Lazy-load Unicode11Addon** (Low risk)
+- Only load when CJK content is detected in terminal output
+- Detection: check for characters in CJK Unicode ranges during ANSI stripping (already iterating)
+- Most sessions never need it
+
+---
+
+## 3. DOM & Session Tab Rendering
+
+### Current State
+- Session tabs use **intelligent incremental updates** with debounced 100ms rendering
+- Incremental path: only updates changed properties (classes, textContent, badges) when session list is stable
+- Full rebuild path: triggered when sessions added/removed **or badge count changes**
+- Subagent windows: per-window xterm.js instances, even when minimized
+
+### Bottlenecks
+
+**B6: Badge count change triggers full tab rebuild** (`app.js:3207-3209`)
+- A single subagent badge increment on one tab triggers `_fullRenderSessionTabs()` — rebuilds entire sidebar HTML via `innerHTML =`
+- With 20 sessions, this is an O(n) reflow for a single badge number change
+- Badge changes are frequent during active subagent work
+
+**B7: Minimized subagent windows retain xterm.js instances** (`subagent-windows.js`)
+- 50 subagent windows × ~75KB per xterm.js instance = ~3.75MB DOM memory
+- Minimized windows are invisible but their terminals remain in DOM
+- xterm.js instances continue processing resize events even when hidden
+
+**B8: `backdrop-filter: blur()` on overlays** (`styles.css:2246-2247, 3098`)
+- Forces new stacking context, disables browser compositing optimizations
+- 50-100ms layout thrashing on modal open/close
+- Only 2 uses, but they're on frequently toggled overlays
+
+### Recommendations
+
+**R6: Targeted badge update without full rebuild** (Low risk)
+- When badge count changes but session list is stable, update only the badge `<span>` textContent
+- Keep incremental path for badge changes; only use full rebuild for structural changes (add/remove sessions)
+- **Savings**: Eliminates O(n) reflow per badge change; reduces to O(1) targeted update
+
+**R7: Lazy xterm.js initialization for subagent windows** (Medium impact)
+- Only create xterm.js Terminal instance when window is restored/maximized
+- On minimize: serialize terminal buffer, dispose Terminal instance, keep buffer in memory
+- On restore: create new Terminal, write buffer back
+- **Savings**: ~3.5MB DOM reduction at 50 minimized agents; eliminates hidden resize processing
+- **Trade-off**: ~200-500ms restore delay (buffer write), mitigated by chunked loading
+
+**R8: Replace `backdrop-filter: blur()` with `background: rgba()`** (Low risk)
+- Use semi-transparent background instead of blur effect
+- Or use `will-change: transform` hint if blur is kept
+- **Savings**: Eliminates forced recomposition layer; 50-100ms faster overlay open
+
+---
+
+## 4. Backend PTY & State Management
+
+### Current State (Excellent)
+- **BufferAccumulator**: Array-based chunking with lazy join on read — avoids O(n) string concatenation
+- **ANSI stripping**: Throttled at 150ms intervals with lazy evaluation (not per-chunk)
+- **State persistence**: 500ms debounce + incremental JSON caching per session (only dirty sessions re-serialized)
+- **Expensive parsers**: Throttled to 150ms window, accumulated data capped at 64KB
+- **Memory**: All buffers have hard limits (2MB terminal, 1MB text, 1000 messages, 64KB line buffer)
+
+### Bottlenecks
+
+**B9: Pending clean data cap at 64KB** (`session.ts:1097-1133`)
+- Between 150ms processing windows, raw PTY data accumulates in `_pendingCleanData`
+- Capped at 64KB — excess data rolls off (old data discarded)
+- During heavy output (large build logs), this means parsers may miss content
+- Acceptable trade-off for performance, but worth documenting
+
+**B10: `LRUMap.delete()` is O(n) worst case** (`utils/lru-map.ts:137-138`)
+- When deleting the newest entry, iterates all keys to find new newest
+- Rare in practice (delete is uncommon; set/get are hot paths)
+- Could matter during mass cleanup of 500 agents
+
+### Recommendations
+
+**R9: Consider adaptive pending data cap** (Low priority)
+- During idle detection (critical to get right), increase cap to 128KB
+- During active working state, keep at 64KB (parsers less critical)
+- **Benefit**: More accurate idle detection during heavy output
+
+**R10: Track second-newest in LRUMap** (Low priority)
+- Maintain a `_secondNewestKey` alongside `_newestKey`
+- On delete of newest, promote second-newest without iteration
+- Only matters at scale (500+ agents with frequent eviction)
+
+---
+
+## 5. Local Echo & Input Path
+
+### Current State (Well-Designed)
+- **DOM overlay approach** — `<span>` elements in `.xterm-screen` at z-index 7, completely independent of `terminal.write()`
+- **Render caching**: `_lastRenderKey` includes text, position, column offsets — skips redundant re-renders
+- **Input flow**: Char accumulation → Enter triggers flush → 80ms delay before `\r` (ensures text reaches PTY first)
+- **Tab completion**: Baseline snapshot → detect buffer change → 300ms fallback timer
+- **CJK support**: Per-character width detection with `terminal.unicode.getStringCellWidth()` preferred, manual fallback
+- **Prompt detection**: Bottom-up line scan, O(rows) — cached position, column-lock prevents jitter
+
+### Bottlenecks
+
+**B11: tmux send-keys latency** (~50-100ms per input)
+- Each `writeViaMux()` spawns a child process (`tmux send-keys`)
+- Text and Enter sent separately with 50ms delay between
+- For rapid typing: characters batch before Enter, so overhead is per-command not per-keystroke
+- **Acceptable trade-off** for session persistence (tmux survives server restarts)
+
+**B12: 80ms delay between text flush and Enter** (`app.js:872-875`)
+- Intentional: ensures text reaches PTY before Enter, preventing Ink from processing empty input
+- Adds 80ms to perceived Enter-to-response latency
+- Could potentially be reduced with acknowledgment-based approach
+
+**B13: Scroll listener on terminal viewport** (`zerolag-input-addon.ts:139`)
+- 50ms debounced re-render on scroll — acceptable but fires frequently during heavy output
+- Overlay hidden when scrolled up (correct behavior), shown when at bottom
+
+### Recommendations
+
+**R11: Reduce Enter delay from 80ms to 50ms** (Low risk, test carefully)
+- The tmux `send-keys` already has 50ms internal delay
+- Combined with network latency, 80ms client-side may be excessive
+- Test with Ink-heavy sessions (Claude Code's status bar) — if text arrives before Enter at 50ms, reduce
+- **Savings**: 30ms perceived latency reduction per command
+
+**R12: Batch tmux send-keys via stdin pipe** (Medium effort, high impact for rapid input)
+- Instead of spawning `tmux send-keys` per input, maintain a persistent connection
+- Use `tmux -C` (control mode) for programmatic interaction without child process spawning
+- **Savings**: Eliminate ~50-100ms process spawn overhead per input
+- **Risk**: Control mode has different semantics; needs careful testing with session persistence
+
+**R13: Skip overlay re-render during heavy output scroll** (Low risk)
+- When terminal is receiving >10KB/s output, hide overlay entirely (user isn't typing during heavy output)
+- Re-show overlay after 500ms of output silence
+- **Savings**: Eliminates unnecessary DOM overlay re-renders during build logs / test output
+
+---
+
+## 6. Polling & File Watchers
+
+### Current State
+- **SubagentWatcher**: 1s base poll, full scan throttled to every 5s, fs.watch() on known directories
+- **TranscriptWatcher**: 1 per session, fs.watch() primary with 1s poll fallback
+- **ImageWatcher**: chokidar per session with 100ms stability poll, burst limit 20/10s
+- **TeamWatcher**: chokidar primary with 30s poll fallback, LRU caches (50 teams, 200 tasks)
+- **RalphTracker**: Todo cleanup every 5 minutes
+
+### Scaling Profile (20 sessions)
+| Component | Instances | Frequency | Total ops/sec |
+|-----------|-----------|-----------|---------------|
+| SubagentWatcher | 1 (global) | Full scan every 5s | 0.2/s |
+| TranscriptWatcher | 20 | 1s poll (fallback) | 20/s max |
+| ImageWatcher | 20 | 100ms poll (during writes only) | 200/s burst |
+| TeamWatcher | 1 (global) | 30s poll (fallback) | 0.03/s |
+| SSE heartbeat | 1 (global) | 15s | 0.07/s |
+| SSE dead client check | 1 (global) | 30s | 0.03/s |
+| Mux stats collection | 1 (global) | 2s | 0.5/s |
+| **Total steady-state** | | | **~21/s** |
+
+### Recommendations
+
+**R14: Increase TranscriptWatcher poll interval to 2s** (Low risk)
+- Transcript changes are infrequent (new messages every few seconds at most)
+- fs.watch() is the primary mechanism; polling is fallback
+- **Savings**: Halves fallback filesystem checks (20/s → 10/s for 20 sessions)
+
+**R15: Share chokidar instances for co-located session directories** (Medium effort)
+- Sessions in the same parent directory could share a single chokidar watcher with depth:3
+- Common case: multiple sessions in `~/projects/foo/` — one watcher covers all
+- **Savings**: Reduce chokidar instances from 20 to ~5-10 for typical workloads
+
+---
+
+## 7. Frontend Asset Delivery
+
+### Current State
+- **app.js**: 12,027 lines (source) → esbuild minified → gzip/brotli compressed (~30-40KB gzipped)
+- **Static caching**: `maxAge: '1y'` via `@fastify/static`
+- **Service worker**: Push notification handler only — no asset caching
+- **No code splitting**: Single monolithic app.js bundle
+
+### Bottlenecks
+
+**B14: No cache-busting mechanism**
+- `maxAge: '1y'` means browsers cache aggressively
+- After deployment, users need `Ctrl+Shift+R` to see updates
+- No content hash in filenames or ETags for automatic invalidation
+
+**B15: Monolithic app.js**
+- All 12K lines loaded on initial page load regardless of which features are used
+- Ralph wizard, plan orchestrator UI, team management — all loaded upfront
+- Mobile loads the same bundle as desktop
+
+### Recommendations
+
+**R16: Add content hash to asset filenames** (Medium impact)
+- Build step: rename `app.js` → `app.[hash].js`
+- Generate a manifest or inject hash into HTML template
+- Keep `maxAge: '1y'` — cache invalidation happens via filename change
+- **Savings**: Eliminates stale cache issues after deployment; removes need for manual hard refresh
+
+**R17: Code-split app.js into core + feature modules** (High effort, medium impact)
+- Core (~4K lines): terminal, SSE, session management, tabs, input handling
+- Deferred (~8K lines): Ralph wizard, plan UI, team management, subagent windows, image viewer
+- Load deferred modules on first use via dynamic `import()` or lazy `<script>` injection
+- **Savings**: ~60% reduction in initial load size; faster time-to-interactive
+- **Risk**: Complexity increase; need to handle loading states for deferred features
+- **Note**: May not be worth the effort given the app is already gzipped to ~30-40KB
+
+---
+
+## 8. CSS Performance
+
+### Current State
+- **styles.css**: 7,153 lines with ~45 box-shadow uses, 2 backdrop-filter uses
+- Animations: GPU-accelerated keyframes for pulsing alerts, loading spinners
+- Z-index layering: well-organized (subagent 1000, plan 1100, log 2000, image 3000, overlay 7)
+
+### Recommendations
+
+**R18: Replace backdrop-filter with opaque overlay** (Low risk, covered in R8)
+
+**R19: Use `contain: content` on subagent windows** (Low risk)
+- Add CSS containment to subagent window containers
+- Prevents layout changes inside windows from triggering reflow on parent
+- Especially valuable with 50 windows: changes in one window won't invalidate others
+- ```css
+  .subagent-window { contain: content; }
+  ```
+- **Savings**: Reduces layout recalculation scope from global to per-window
+
+**R20: Use `content-visibility: auto` on off-screen subagent windows** (Low risk)
+- Browser skips rendering of off-screen windows entirely
+- Combined with `contain-intrinsic-size` to prevent layout shift
+- ```css
+  .subagent-window.minimized { content-visibility: hidden; }
+  ```
+- **Savings**: Browser skips paint/layout for minimized windows; complements R7
+
+---
+
+## 9. Memory & Scaling Limits
+
+### Current Budget (20 sessions)
+| Component | Per Session | Total | Status |
+|-----------|-----------|-------|--------|
+| Terminal buffer | 2MB | 40MB | Hard-limited, auto-trim |
+| Text output | 1MB | 20MB | Hard-limited, auto-trim |
+| Messages | ~1MB | 20MB | Capped at 1000, trims to 800 |
+| Respawn buffer | 1MB | 20MB | Hard-limited |
+| **Buffers total** | | **100MB** | Acceptable |
+| TranscriptWatcher | ~100KB | 2MB | |
+| ImageWatcher | ~50KB | 1MB | |
+| SubagentWatcher | ~500KB | 500KB | Global |
+| Frontend terminal cache | ~256KB | 5MB | LRU, max 20 entries |
+| **Total estimated** | | **~110MB** | Comfortable |
+
+### At Max Scale (50 sessions)
+- Buffers: ~250MB
+- Watchers: ~5MB
+- **Total: ~255MB** + Node.js overhead — acceptable on modern hardware
+
+### Potential Leak Vectors (All Mitigated)
+- `_shortIdCache` in server — unbounded Map, but entries are tiny (string→string); grows at O(sessions created), not O(events)
+- All CleanupManager-registered resources tracked and disposed on session stop
+- `isStopped` guard prevents new timers after session cleanup
+
+---
+
+## 10. Implementation Priority Matrix
+
+### Phase 1 — Quick Wins (1-2 hours each, low risk)
+| # | Optimization | Files to Change |
+|---|-------------|-----------------|
+| R6 | Targeted badge update | `app.js` (3207-3209) |
+| R3 | Canvas renderer on mobile | `app.js` (627-637) |
+| R8 | Replace backdrop-filter blur | `styles.css` (2246, 3098) |
+| R19 | CSS containment on subagent windows | `styles.css` |
+| R20 | `content-visibility: hidden` on minimized windows | `styles.css` |
+
+### Phase 2 — Medium Effort (half-day each)
+| # | Optimization | Files to Change |
+|---|-------------|-----------------|
+| R2 | Tiered SSE padding | `server.ts` (broadcast function) |
+| R7 | Lazy xterm.js for minimized subagents | `subagent-windows.js` |
+| R11 | Reduce Enter delay to 50ms | `app.js` (872-875), test with Ink |
+| R14 | TranscriptWatcher 2s poll | `transcript-watcher.ts` |
+| R16 | Content-hash asset filenames | `build.mjs`, `server.ts` |
+
+### Phase 3 — Larger Initiatives (1-2 days each)
+| # | Optimization | Files to Change |
+|---|-------------|-----------------|
+| R1 | Session-scoped SSE subscriptions | `server.ts`, `app.js` (SSE connect) |
+| R5 | Lazy Unicode11Addon loading | `app.js`, build pipeline |
+| R12 | Persistent tmux control mode | `tmux-manager.ts` |
+| R17 | Code-split app.js | `app.js`, `build.mjs`, HTML template |
+
+### Not Recommended (Low ROI or High Risk)
+| # | Why Not |
+|---|---------|
+| R4 | Dynamic scrollback adds complexity; memory savings marginal vs total budget |
+| R9 | Adaptive pending data cap adds state; current 64KB cap rarely matters |
+| R10 | LRUMap.delete() O(n) is theoretical; never triggered at current scale |
+| R15 | Shared chokidar instances add directory-matching complexity for minimal gain |
+
+---
+
+## Appendix: Key File Locations
+
+| Area | File | Key Lines |
+|------|------|-----------|
+| SSE broadcast | `src/web/server.ts` | 1961-1989 (broadcast), 1934-1959 (backpressure) |
+| Terminal batching | `src/web/server.ts` | 1994-2048 (per-session adaptive batching) |
+| Frame budget | `src/web/public/app.js` | 1370-1478 (flushPendingWrites, 64KB cap) |
+| Flicker filter | `src/web/public/app.js` | 1176-1255 (50ms sync wait, 256KB safety) |
+| Tab rendering | `src/web/public/app.js` | 3108-3357 (incremental + full rebuild) |
+| Tab switching | `src/web/public/app.js` | 3560-3760 (cache + chunked load + deferred UI) |
+| Local echo | `packages/xterm-zerolag-input/src/` | All files (overlay, prompt, CJK) |
+| Local echo integration | `src/web/public/app.js` | 640, 815-988 (input flow) |
+| Subagent windows | `src/web/public/subagent-windows.js` | Full file (window mgmt, drag, minimize) |
+| State persistence | `src/state-store.ts` | 161-250 (debounced save, incremental JSON) |
+| Buffer accumulator | `src/utils/buffer-accumulator.ts` | Full file (array chunks, lazy join) |
+| PTY handling | `src/session.ts` | 1046-1133 (data flow), 1173-1230 (parsing) |
+| Config limits | `src/config/` | 9 files (buffer, map, timing, auth, etc.) |
+| Anti-flicker docs | `docs/terminal-anti-flicker.md` | Architecture reference |
+| CSS | `src/web/public/styles.css` | 2246 (backdrop-filter), full file |
+| Build pipeline | `scripts/build.mjs` | 59-68 (minify + compress) |

docs/performance-optimization-plan.md

@@ -1,7 +1,7 @@
 # Performance & Responsiveness Optimization Plan
 
 **Date**: 2026-02-28
-**Status**: In Progress
+**Status**: Phases 1–4 Complete. Phase 5 optional/deferred.
 
 ---
 
@@ -13,7 +13,7 @@ Three independent research passes analyzed the Codeman codebase for performance
 
 ---
 
-## Phase 1: Quick Wins — ALREADY IMPLEMENTED
+## Phase 1: Quick Wins — COMPLETE
 
 All Phase 1 items were found to already exist in the codebase during verification:
 
@@ -27,7 +27,7 @@ All Phase 1 items were found to already exist in the codebase during verificatio
 
 ---
 
-## Phase 2: Frontend Responsiveness — MOSTLY ALREADY IMPLEMENTED
+## Phase 2: Frontend Responsiveness — COMPLETE
 
 ### 2.1 Batch `getBoundingClientRect()` in connection lines — DONE
 - **Files**: `src/web/public/app.js` (`_updateConnectionLinesImmediate()`)
@@ -51,7 +51,7 @@ All Phase 1 items were found to already exist in the codebase during verificatio
 
 ---
 
-## Phase 3: Backend Hot Paths
+## Phase 3: Backend Hot Paths — COMPLETE
 
 ### 3.1 State diff broadcasts — ALREADY OPTIMIZED
 - `broadcastSessionStateDebounced()` already batches at 500ms intervals
@@ -84,35 +84,37 @@ All Phase 1 items were found to already exist in the codebase during verificatio
 
 ---
 
-## Phase 4: System-Level Improvements
+## Phase 4: System-Level Improvements — COMPLETE
 
-### 4.1 Incremental state persistence
-- **Files**: `src/state-store.ts` (~lines 145-160)
-- **Problem**: Every 500ms debounce writes the entire `AppState` (all sessions, tasks, config) via `JSON.stringify()`. With 50 sessions, state can be tens of MB. Serialization alone costs 50-100ms.
-- **Fix**: Track dirty sessions. On persist, only re-serialize dirty sessions; cache serialized JSON for clean sessions. Assemble final output from cached fragments.
-- **Impact**: Reduces serialization cost from O(all sessions) to O(dirty sessions). Typical steady-state: 1-2 dirty sessions instead of 50.
+### 4.1 Incremental state persistence — DONE
+- **Files**: `src/state-store.ts` (`assembleStateJson()`, `setSession()`)
+- **Change**: Added `dirtySessions` Set and `cachedSessionJsons` Map. On persist, only dirty sessions are re-serialized; clean sessions reuse cached JSON fragments. `setSession()` marks sessions dirty; `assembleStateJson()` rebuilds only changed fragments.
+- **Impact**: Serialization cost reduced from O(all sessions) to O(dirty sessions). Typical steady-state: 1-2 dirty sessions instead of 50.
 
-### 4.2 Replace polling with fs watchers for team watcher
-- **Files**: `src/team-watcher.ts` (~lines 148-180)
-- **Problem**: Polls `~/.claude/teams/` every 5s via `readdir()` + `stat()`. Blocks event loop for 100-200ms on large directories.
-- **Fix**: Use `chokidar` (already a dependency) or `fs.watch()` to react to changes. Keep a 30s fallback poll for reliability.
-- **Impact**: Eliminates 5s polling overhead; near-instant team detection.
+### 4.2 Replace polling with fs watchers for team watcher — DONE
+- **Files**: `src/team-watcher.ts` (`setupFsWatchers()`)
+- **Change**: Added chokidar watchers on both `~/.claude/teams/` and `~/.claude/tasks/` directories for instant event-driven detection. Lock files ignored via chokidar config. Mtime-based dedup skips unchanged files. Polling interval relaxed from 5s to 30s as a fallback.
+- **Impact**: Near-instant team detection; polling overhead eliminated for normal operation.
 
-### 4.3 Consolidate subagent file watchers
-- **Files**: `src/subagent-watcher.ts` (~line 229+)
-- **Problem**: One chokidar watcher per agent directory. With 500 agents, that's 500 inotify watchers consuming kernel resources.
-- **Fix**: Watch at the session level (one watcher per session's subagent directory), not per-agent. Parse events to route to correct agent.
-- **Impact**: Reduces inotify watchers from 500 to ~50 (one per session).
+### 4.3 Consolidate subagent file watchers — DONE
+- **Files**: `src/subagent-watcher.ts` (`setupDirectoryWatcher()`)
+- **Change**: Replaced per-agent chokidar watchers with one `fs.watch()` per session subagent directory. Events are routed to the correct agent via filename. Per-file debouncing (100ms) prevents hammering on bulk discovery.
+- **Impact**: Inotify watchers reduced from potentially 500 (one per agent) to ~50 (one per session directory).
 
-### 4.4 Stream transcript files instead of full reads
-- **Files**: `src/subagent-watcher.ts` (~lines 959-964)
-- **Problem**: `loadTranscript()` reads entire transcript file (can be >100KB). With 500 agents discovered at once, that's 50MB of file reads.
-- **Fix**: Only read last 10KB for display (tail). Full file on-demand only (e.g., when user opens transcript viewer).
-- **Impact**: Reduces file I/O from 50MB to 5MB for bulk agent discovery.
+### 4.4 Stream transcript files instead of full reads — DONE
+- **Files**: `src/subagent-watcher.ts` (`tailFile()`, `findDescriptionInAgentFile()`, parent transcript lookup)
+- **Change**: Multiple streaming strategies implemented:
+  - **Live monitoring**: Position-based `tailFile()` with `createReadStream({ start: fromPosition })` — only reads new content
+  - **Parent transcript lookup**: Streams only last 16KB (`createReadStream({ start: offset })`)
+  - **Description extraction**: Streams only first 8KB, exits early after 5 lines
+  - **Full read**: Only for on-demand transcript review panel (with optional `limit` parameter)
+- **Impact**: File I/O for bulk agent discovery reduced from ~50MB to ~5MB.
 
 ---
 
-## Phase 5: Long-Term Architectural (Optional)
+## Phase 5: Long-Term Architectural (Optional) — NOT STARTED
+
+These items are deferred until scaling demands justify the complexity.
 
 ### 5.1 Worker thread for PTY processing
 - **Files**: `src/session.ts`
@@ -134,42 +136,23 @@ All Phase 1 items were found to already exist in the codebase during verificatio
 
 ---
 
-## Priority Matrix (Remaining Work)
+## Completion Summary
 
-| # | Item | Impact | Risk | Effort |
-|---|------|--------|------|--------|
-| 3.1 | State diff broadcasts | **Very High** | Medium | 3-4h |
-| 3.2 | Fix session cache invalidation | **High** | Low | 1h |
-| 3.3 | Skip PTY processing for hidden sessions | **High** | Medium | 2-3h |
-| 3.5 | Throttle detection broadcasts | **Medium** | Low | 1h |
-| 3.4 | Batch liveness checks | **Medium** | Low | 1-2h |
-| 4.1 | Incremental state persistence | **Medium** | Medium | 3-4h |
-| 4.2 | Team watcher fs events | **Low-Med** | Medium | 2h |
-| 4.3 | Consolidate file watchers | **Low-Med** | Medium | 2h |
-| 4.4 | Stream transcripts | **Low-Med** | Low | 1h |
-| 5.1 | Worker thread PTY | **Med** (at scale) | High | 8h |
-| 5.2 | Per-session SSE subs | **Med** (at scale) | High | 4h |
-| 5.3 | O(1) LRUMap | **Very Low** | Medium | 2h |
+| Phase | Scope | Status | Items |
+|-------|-------|--------|-------|
+| 1 | Quick Wins | **Complete** | 5/5 (all pre-existing) |
+| 2 | Frontend Responsiveness | **Complete** | 3/3 actionable done, 2 skipped |
+| 3 | Backend Hot Paths | **Complete** | 4/4 actionable done, 1 deferred |
+| 4 | System-Level | **Complete** | 4/4 done |
+| 5 | Long-Term Architectural | **Not started** | 0/3 — deferred until needed |
 
----
-
-## Recommended Execution Order
-
-**Sprint 1** (Phase 3 — Backend Hot Paths): Items 3.1, 3.2, 3.3, 3.5
-- Backend serialization and broadcast efficiency
-- Highest remaining impact; requires careful testing with multiple active sessions
-
-**Sprint 2** (Phase 4 — System Level): Items 4.1, 3.4, 4.3, 4.4
-- State persistence, liveness checks, watcher consolidation
-- Medium-complexity refactors
-
-**Sprint 3** (Phase 5 — Architectural): Items 5.1, 5.2 — only if scaling demands it
+**Overall**: 16/16 actionable items complete. 3 optional items deferred.
 
 ---
 
 ## Measurement
 
-Before starting implementation, establish baselines:
+Before starting Phase 5, establish baselines:
 
 1. **Frontend**: Record Chrome DevTools Performance trace with 10 sessions open. Measure:
    - Frame rate during rapid terminal output

docs/performance-plan.md

@@ -0,0 +1,74 @@
+# Codeman Performance Optimization Plan
+
+## Current State
+
+The backend is **already production-grade** — SSE broadcasting, state persistence, terminal batching, buffer management, and memory patterns are all well-optimized. The biggest gains are on the **frontend delivery** side.
+
+## Implemented Optimizations
+
+### 1. V8 Compile Cache (10-20% faster cold start)
+
+**Files:** `scripts/codeman-web.service`, `package.json`
+
+Node.js re-parses and compiles all JS on every cold start. `NODE_COMPILE_CACHE` caches V8 compiled bytecode to disk, reusing it on subsequent starts.
+
+- Added `Environment=NODE_COMPILE_CACHE=/home/arkon/.codeman/compile-cache` to systemd service
+- Added to `npm start` script for non-systemd usage
+- Zero code changes, immediate win on every restart
+
+### 2. WebGL Addon Lazy-Loading (244KB saved on mobile, non-blocking on desktop)
+
+**Files:** `src/web/public/index.html`, `src/web/public/app.js`
+
+`xterm-addon-webgl.min.js` (244KB) was loaded eagerly for all users via `<script defer>`, but only used on desktop with WebGL2 support.
+
+- Removed `<script defer>` from `index.html`
+- Added dynamic script loading in `app.js` — only downloads on desktop when WebGL is needed
+- Mobile users never download the file at all (244KB saved)
+- Desktop: loads in parallel with page rendering, addon initializes when ready
+- Graceful fallback: canvas renderer used if WebGL unavailable or script fails
+
+### 3. Preload Hints (~50-100ms faster perceived load)
+
+**Files:** `src/web/public/index.html`
+
+Browser discovers `<script defer>` tags only when the parser reaches them at the bottom of `<body>`. By then, the HTML parse has blocked for hundreds of lines.
+
+- Added `<link rel="preload" as="script">` in `<head>` for `vendor/xterm.min.js`, `constants.js`, `app.js`
+- Browser starts fetching critical scripts immediately during HTML parse (before reaching `<body>`)
+- Zero runtime overhead — just hints for the browser's preload scanner
+
+### 4. Batch Tmux Reconciliation (N subprocess calls → 1)
+
+**Files:** `src/tmux-manager.ts`
+
+`reconcileSessions()` previously called `tmux has-session` + `tmux display-message` per known session, plus `tmux list-sessions` for discovery, plus `tmux display-message` per discovered session. With 20 sessions: 41+ subprocess calls.
+
+- Replaced with single `tmux list-panes -a -F '#{session_name}\t#{pane_pid}'` call
+- Builds a Map from the result, then does O(1) lookups for both known and discovered sessions
+- Also replaced inner O(n) `isKnown` scan with a Set lookup
+- 20 sessions: 41 subprocess calls → 1, with faster lookups
+
+### 5. Asset Hashing / Cache Busting (already implemented)
+
+**Files:** `scripts/build.mjs` (pre-existing)
+
+Content-hash cache busting was already implemented in the build script:
+- All app JS/CSS files get content hashes (`app.abc123.js`)
+- `index.html` rewritten to reference hashed filenames
+- Pre-compressed with gzip + Brotli
+- 1-year immutable cache works correctly — new deploys get new filenames
+
+## Already Optimized (No Action Needed)
+
+| Area | Why It's Fine |
+|------|---------------|
+| **SSE Broadcasting** | Single serialization per broadcast, preformatted frames, backpressure handling, session subscription filtering |
+| **State Persistence** | 500ms debounce, incremental per-session JSON caching, async atomic writes, circuit breaker on failures |
+| **Terminal Batching** | Adaptive intervals (16-50ms), per-session queues, immediate flush at 32KB, array-based accumulation |
+| **Buffer Management** | BufferAccumulator (array-push, lazy join), auto-trim at 2MB/1MB, no string concatenation in hot paths |
+| **ANSI Stripping** | Pre-compiled regex via factory functions, single-pass processing |
+| **Static File Serving** | @fastify/static with 1-year cache, pre-compressed Brotli/gzip, no-cache for HTML |
+| **Memory Management** | CleanupManager, LRUMap, StaleExpirationMap, bounded buffers, explicit listener cleanup |
+| **Import Patterns** | Pure ESM, lazy web server import, no circular deps, no dynamic imports in hot paths |
+| **Config Loading** | Small constant files, no I/O at import time, specific imports (no barrel) |

docs/phase-4-domain-splitting-plan.md

@@ -0,0 +1,788 @@
+# Phase 4: Domain File Splitting — Implementation Plan
+
+**Date**: 2026-03-01
+**Prerequisites**: Phase 1-3 complete (utils cleanup, CleanupManager/Debouncer migration, route extraction)
+**Goal**: Split 4 god files into focused modules with barrel exports for transparent migration.
+
+---
+
+## Table of Contents
+
+1. [Split types.ts into types/ directory](#1-split-typests-into-types-directory)
+2. [Split ralph-tracker.ts into focused modules](#2-split-ralph-trackerts-into-focused-modules)
+3. [Split respawn-controller.ts into focused modules](#3-split-respawn-controllerts-into-focused-modules)
+4. [Split session.ts into focused modules](#4-split-sessionts-into-focused-modules)
+5. [Execution Order & Dependencies](#5-execution-order--dependencies)
+6. [Validation Checklist](#6-validation-checklist)
+
+---
+
+## 1. Split types.ts into types/ directory
+
+**Current**: 1,443 lines, 71 exports, imported by 36 files.
+**Risk**: LOW — pure type refactor, no runtime behavior change.
+
+### Target Structure
+
+```
+src/types/
+├── index.ts              (barrel re-export — transparent migration)
+├── common.ts             (Disposable, BufferConfig, CleanupResourceType, CleanupRegistration)
+├── session.ts            (SessionStatus, SessionMode, ClaudeMode, SessionConfig, SessionColor,
+│                          SessionState, OpenCodeConfig, SessionOutput)
+├── task.ts               (TaskStatus, TaskDefinition, TaskState)
+├── app-state.ts          (AppState, AppConfig, GlobalStats, TokenUsageEntry, TokenStats,
+│                          DEFAULT_CONFIG, createInitialState, createInitialGlobalStats)
+├── respawn.ts            (RespawnConfig, PersistedRespawnConfig, CycleOutcome,
+│                          RespawnCycleMetrics, RespawnAggregateMetrics, HealthStatus,
+│                          RalphLoopHealthScore, TimingHistory, RespawnPreset)
+├── ralph.ts              (RalphLoopStatus, RalphLoopState, RalphTodoStatus, RalphTodoPriority,
+│                          RalphTodoItem, RalphTodoProgress, RalphSessionState,
+│                          RalphStatusValue, RalphTestsStatus, RalphWorkType, RalphStatusBlock,
+│                          CompletionConfidence, RalphTrackerState,
+│                          CircuitBreakerState, CircuitBreakerReason, CircuitBreakerStatus,
+│                          createInitialCircuitBreakerStatus, createInitialRalphTrackerState,
+│                          createInitialRalphSessionState)
+├── api.ts                (ApiErrorCode, ApiResponse, HookEventType, QuickStartResponse,
+│                          CaseInfo, createErrorResponse, isError, getErrorMessage)
+├── lifecycle.ts          (LifecycleEventType, LifecycleEntry)
+├── run-summary.ts        (RunSummaryEventType, RunSummaryEventSeverity, RunSummaryEvent,
+│                          RunSummaryStats, RunSummary, createInitialRunSummaryStats)
+├── tools.ts              (ActiveBashToolStatus, ActiveBashTool, ImageDetectedEvent)
+├── teams.ts              (TeamConfig, TeamMember, TeamTask, InboxMessage, PaneInfo)
+├── push.ts               (PushSubscriptionRecord, VapidKeys)
+└── plan.ts               (PlanTaskStatus, TddPhase, PlanItem re-export, NiceConfig,
+                            DEFAULT_NICE_CONFIG, ProcessStats)
+```
+
+### Steps
+
+1. **Create `src/types/` directory** and each domain file above.
+
+2. **Move types** from `src/types.ts` into their domain files. Preserve all JSDoc comments. Each file should import from siblings as needed (e.g., `ralph.ts` imports `CircuitBreakerState` within itself — no cross-file deps needed since they're in the same file).
+
+3. **Create barrel `src/types/index.ts`** that re-exports everything:
+   ```typescript
+   export * from './common.js';
+   export * from './session.js';
+   export * from './task.js';
+   export * from './app-state.js';
+   export * from './respawn.js';
+   export * from './ralph.js';
+   export * from './api.js';
+   export * from './lifecycle.js';
+   export * from './run-summary.js';
+   export * from './tools.js';
+   export * from './teams.js';
+   export * from './push.js';
+   export * from './plan.js';
+   ```
+
+4. **Delete old `src/types.ts`** and replace with a single-line re-export barrel:
+   ```typescript
+   export * from './types/index.js';
+   ```
+   This ensures `import { ... } from './types.js'` continues to work everywhere — zero changes to 36 import sites.
+
+5. **Verify**: `tsc --noEmit` and `npm run lint` must pass. No runtime changes.
+
+### Internal Dependencies Between Domain Files
+
+Some types reference others across domains. Handle with imports:
+
+| File | Imports From |
+|------|-------------|
+| `app-state.ts` | `session.ts` (SessionState), `task.ts` (TaskState), `ralph.ts` (RalphLoopState, RalphSessionState) |
+| `respawn.ts` | None (self-contained) |
+| `ralph.ts` | None (self-contained) |
+| `run-summary.ts` | None (self-contained) |
+| `api.ts` | None (self-contained) |
+| `session.ts` | `respawn.ts` (RespawnConfig), `ralph.ts` (RalphTrackerState, RalphTodoItem, CircuitBreakerStatus, RalphSessionState, RunSummaryEvent) |
+
+Wait — `SessionState` references `RespawnConfig`, `RalphTrackerState`, `CircuitBreakerStatus`, and `RunSummaryEvent`. This creates imports from `session.ts` → `respawn.ts`, `ralph.ts`, `run-summary.ts`. This is fine (one-way deps, no cycles).
+
+---
+
+## 2. Split ralph-tracker.ts into focused modules
+
+**Current**: 3,868 lines, single `RalphTracker` class with 5 responsibilities.
+**Risk**: MEDIUM — class has shared mutable state, but extractable modules are well-isolated.
+
+### Coupling Analysis Summary
+
+| Module | Coupling | Extractability |
+|--------|----------|----------------|
+| Plan task tracking | LOW | HIGH — only reads `cycleCount` |
+| Fix-plan file watching | LOW | HIGH — callback-based todo replacement |
+| Iteration stall detection | LOW | HIGH — notification-based |
+| RALPH_STATUS block parsing + circuit breaker | MEDIUM | MEDIUM — callback for circuit breaker updates |
+| Todo parsing, loop detection, completion | HIGH | LOW — deeply entangled shared state |
+
+### Target Structure
+
+```
+src/
+├── ralph-tracker.ts              (~1,800 LOC — core: output parsing, loop state,
+│                                   todo management, completion detection)
+├── ralph-plan-tracker.ts         (~600 LOC — plan tasks, checkpoints, history, rollback)
+├── ralph-status-parser.ts        (~300 LOC — RALPH_STATUS block parsing, circuit breaker)
+├── ralph-fix-plan-watcher.ts     (~150 LOC — @fix_plan.md file watching)
+└── ralph-stall-detector.ts       (~80 LOC — iteration stall detection)
+```
+
+### Step 2a: Extract `RalphPlanTracker` (~600 LOC)
+
+**Why first**: Lowest coupling. Only dependency is `cycleCount` for checkpoint detection.
+
+**Extract these from `RalphTracker`**:
+
+Types to export:
+- `EnhancedPlanTask` (interface, currently lines 56-87)
+- `CheckpointReview` (interface, currently lines 90-139)
+
+Properties to move:
+- `_planVersion: number`
+- `_planHistory: Array<{version, timestamp, tasks, summary}>`
+- `_planTasks: Map<string, EnhancedPlanTask>`
+- `_checkpointIterations: number[]`
+- `_lastCheckpointIteration: number`
+
+Methods to move:
+- `initializePlanTasks(items)`
+- `updatePlanTask(taskId, update)`
+- `addPlanTask(params)`
+- `getPlanTasks()`
+- `generateCheckpointReview()`
+- `getPlanHistory()`
+- `rollbackToVersion(version)`
+- `isCheckpointDue()`
+- `planVersion` getter
+- `_savePlanToHistory()` (private)
+- `_unblockDependentTasks()` (private)
+- `_checkForCheckpoint()` (private)
+
+Events emitted (define in new class):
+- `planInitialized`
+- `planTaskUpdate`
+- `taskBlocked`
+- `taskUnblocked`
+- `planCheckpoint`
+
+**Interface with parent**:
+```typescript
+export class RalphPlanTracker extends EventEmitter {
+  constructor() { ... }
+
+  // Parent calls this when iteration changes (for checkpoint detection)
+  notifyCycleCount(cycleCount: number): void { ... }
+
+  // Full public API moves here unchanged
+  initializePlanTasks(items: PlanItem[]): void { ... }
+  updatePlanTask(taskId: string, update: { ... }): { ... } | null { ... }
+  // ...etc
+}
+```
+
+**In `RalphTracker`**: Replace plan methods with delegation:
+```typescript
+readonly planTracker = new RalphPlanTracker();
+
+// Forward plan events
+this.planTracker.on('planInitialized', (...args) => this.emit('planInitialized', ...args));
+// ...etc
+
+// In detectLoopStatus(), when cycleCount changes:
+this.planTracker.notifyCycleCount(this._loopState.cycleCount);
+```
+
+### Step 2b: Extract `RalphFixPlanWatcher` (~150 LOC)
+
+**Extract these**:
+
+Properties:
+- `_workingDir: string | null`
+- `_fixPlanPath: string | null`
+- `_fixPlanWatcher: FSWatcher | null`
+- `_fixPlanWatcherErrorHandler`
+- `_fixPlanReloadDeb`
+
+Methods:
+- `setWorkingDir(workingDir)`
+- `loadFixPlanFromDisk()`
+- `startWatchingFixPlan()`
+- `stopWatchingFixPlan()`
+- `handleFixPlanChange()`
+- `isFileAuthoritative` getter
+
+**Interface with parent**:
+```typescript
+export class RalphFixPlanWatcher extends EventEmitter {
+  get isFileAuthoritative(): boolean { ... }
+
+  setWorkingDir(workingDir: string): void { ... }
+  stop(): void { ... }
+}
+
+// Events:
+// 'todosLoaded' → (todos: Array<{id, content, status, priority}>) — parent replaces _todos
+```
+
+**In `RalphTracker`**:
+```typescript
+readonly fixPlanWatcher = new RalphFixPlanWatcher();
+
+constructor() {
+  this.fixPlanWatcher.on('todosLoaded', (items) => {
+    // Replace _todos with file-based items
+    this._todos.clear();
+    for (const item of items) {
+      this.addOrUpdateTodo(item.id, item.content, item.status, item.priority);
+    }
+  });
+}
+
+// Delegate isFileAuthoritative
+get isFileAuthoritative(): boolean {
+  return this.fixPlanWatcher.isFileAuthoritative;
+}
+```
+
+### Step 2c: Extract `RalphStallDetector` (~80 LOC)
+
+**Extract these**:
+
+Properties:
+- `_lastIterationChangeTime`
+- `_lastObservedIteration`
+- `_iterationStallTimerId`
+- `_iterationStallWarningMs`
+- `_iterationStallCriticalMs`
+- `_iterationStallWarned`
+
+Methods:
+- `startIterationStallDetection()`
+- `stopIterationStallDetection()`
+- `checkIterationStall()`
+- `getIterationStallMetrics()`
+- `configureIterationStallThresholds(warningMs, criticalMs)`
+
+**Interface with parent**:
+```typescript
+export class RalphStallDetector extends EventEmitter {
+  constructor(private cleanup: CleanupManager) { ... }
+
+  start(): void { ... }
+  stop(): void { ... }
+
+  // Parent calls when iteration changes
+  notifyIterationChanged(iteration: number): void {
+    this._lastIterationChangeTime = Date.now();
+    this._lastObservedIteration = iteration;
+    this._iterationStallWarned = false;
+  }
+
+  // Parent calls to check if loop is active
+  setLoopActive(active: boolean): void { ... }
+
+  getIterationStallMetrics(): { ... } { ... }
+}
+
+// Events: 'iterationStallWarning', 'iterationStallCritical'
+```
+
+### Step 2d: Extract `RalphStatusParser` (~300 LOC)
+
+**Extract these**:
+
+Properties:
+- `_circuitBreaker: CircuitBreakerStatus`
+- `_statusBlockBuffer: string[]`
+- `_inStatusBlock: boolean`
+- `_lastStatusBlock: RalphStatusBlock | null`
+- `_completionIndicators: number`
+- `_exitGateMet: boolean`
+- `_totalFilesModified: number`
+- `_totalTasksCompleted: number`
+
+Methods:
+- `processStatusBlockLine(line)`
+- `parseStatusBlock(lines)`
+- `detectCompletionIndicators(line)`
+- `updateCircuitBreaker(hasProgress, testsStatus, status)`
+- `resetCircuitBreaker()`
+- `circuitBreakerStatus` getter
+- `lastStatusBlock` getter
+- `cumulativeStats` getter
+- `exitGateMet` getter
+
+Regex patterns to move:
+- `RALPH_STATUS_START_PATTERN` through `RALPH_RECOMMENDATION_PATTERN`
+- `COMPLETION_INDICATOR_PATTERNS`
+
+**Interface with parent**:
+```typescript
+export class RalphStatusParser extends EventEmitter {
+  processLine(line: string): void { ... }  // calls processStatusBlockLine + detectCompletionIndicators
+
+  get circuitBreakerStatus(): CircuitBreakerStatus { ... }
+  get lastStatusBlock(): RalphStatusBlock | null { ... }
+  get exitGateMet(): boolean { ... }
+  get cumulativeStats(): { ... } { ... }
+
+  resetCircuitBreaker(): void { ... }
+  reset(): void { ... }
+}
+
+// Events: 'statusBlockDetected', 'circuitBreakerUpdate', 'exitGateMet'
+```
+
+**In `RalphTracker.processLine()`**:
+```typescript
+// Replace inline status block handling with delegation
+this.statusParser.processLine(line);
+```
+
+### Step 2e: Keep in `ralph-tracker.ts` (~1,800 LOC)
+
+The core remains tightly coupled and stays together:
+- Output parsing pipeline (`processTerminalData`, `processCleanData`, `processLine`)
+- Loop state management (`_loopState`, `detectLoopStatus`, `enable/disable/startLoop/stopLoop`)
+- Todo management (`_todos`, `detectTodoItems`, `addOrUpdateTodo`, `updateTodoStatus`, `getTodoStats`)
+- Completion detection (`detectCompletionPhrase`, `handleCompletionPhrase`, `calculateCompletionConfidence`)
+- All-tasks-complete detection (`detectAllTasksComplete`)
+- Auto-enable logic (`shouldAutoEnable`)
+- Lifecycle (`reset`, `fullReset`, `clear`, `restoreState`, `destroy`)
+- Event debouncing and buffering
+
+The class coordinates the extracted modules via composition:
+```typescript
+export class RalphTracker extends EventEmitter {
+  readonly planTracker = new RalphPlanTracker();
+  readonly fixPlanWatcher = new RalphFixPlanWatcher();
+  readonly stallDetector: RalphStallDetector;
+  readonly statusParser = new RalphStatusParser();
+
+  constructor() {
+    super();
+    this.stallDetector = new RalphStallDetector(this.cleanup);
+    this._wireSubModuleEvents();
+  }
+
+  private _wireSubModuleEvents(): void {
+    // Forward all sub-module events through RalphTracker
+    // so external consumers don't need to know about the split
+    for (const event of ['planInitialized', 'planTaskUpdate', ...]) {
+      this.planTracker.on(event, (...args) => this.emit(event, ...args));
+    }
+    // ...same for statusParser, stallDetector, fixPlanWatcher
+  }
+}
+```
+
+### Migration Safety
+
+- All events continue to be emitted from `RalphTracker` (forwarded from sub-modules)
+- All public methods stay on `RalphTracker` (delegated to sub-modules)
+- External consumers (`session.ts`, `case-routes.ts`) see zero API changes
+- New sub-modules are exposed as `readonly` properties for direct access where needed
+
+---
+
+## 3. Split respawn-controller.ts into focused modules
+
+**Current**: 3,611 lines, single `RespawnController` class with 6 responsibilities.
+**Risk**: MEDIUM — health scoring and metrics are cleanly decoupled; detection is tightly coupled.
+
+### Coupling Analysis Summary
+
+| Module | Coupling | Extractability |
+|--------|----------|----------------|
+| Health scoring | NONE | HIGH — pure calculations from metrics |
+| Cycle metrics | LOW | HIGH — standalone tracking |
+| Adaptive timing | LOW | HIGH — standalone timing adjustments |
+| Stuck-state detection | LOW | MEDIUM — needs state + config refs |
+| Pattern detection utilities | NONE | HIGH — pure functions |
+| State machine + idle detection + AI checkers | HIGH | LOW — deeply entangled |
+
+### Target Structure
+
+```
+src/
+├── respawn-controller.ts             (~2,200 LOC — state machine, idle detection,
+│                                       AI checkers, terminal handling, hook signals,
+│                                       auto-accept, step execution)
+├── respawn-health.ts                 (~250 LOC — health scoring + recommendations)
+├── respawn-metrics.ts                (~200 LOC — cycle metrics + aggregate stats)
+├── respawn-adaptive-timing.ts        (~100 LOC — adaptive timing with percentile calc)
+└── respawn-patterns.ts               (~50 LOC — terminal pattern detection utilities)
+```
+
+### Step 3a: Extract `RespawnPatterns` (~50 LOC)
+
+**Pure utility functions, zero coupling**.
+
+Move:
+- `isCompletionMessage(data): boolean`
+- `hasWorkingPattern(data, window): boolean`
+- `extractTokenCount(data): number | null`
+- `PROMPT_PATTERNS` array
+- `WORKING_PATTERNS` array
+
+```typescript
+// src/respawn-patterns.ts
+import { TOKEN_PATTERN, SPINNER_PATTERN } from './utils/index.js';
+
+export const PROMPT_PATTERNS = ['❯', '>', '$', '%', '#'];
+
+export const WORKING_PATTERNS = [/* 70+ patterns */];
+
+export function isCompletionMessage(data: string): boolean { ... }
+export function hasWorkingPattern(data: string, window: string): boolean { ... }
+export function extractTokenCount(data: string): number | null { ... }
+```
+
+**In `RespawnController`**: Import and call:
+```typescript
+import { isCompletionMessage, hasWorkingPattern, extractTokenCount } from './respawn-patterns.js';
+```
+
+### Step 3b: Extract `RespawnAdaptiveTiming` (~100 LOC)
+
+**Self-contained timing controller**.
+
+Move properties:
+- `timingHistory: TimingHistory`
+
+Move methods:
+- `recordTimingData(idleDetectionMs, cycleDurationMs)`
+- `updateAdaptiveTiming()`
+- `getTimingHistory()`
+- `getAdaptiveCompletionConfirmMs()`
+
+```typescript
+export class RespawnAdaptiveTiming {
+  private timingHistory: TimingHistory;
+
+  constructor(private config: { adaptiveMinConfirmMs: number; adaptiveMaxConfirmMs: number }) {
+    this.timingHistory = { recentIdleDetectionMs: [], recentCycleDurationMs: [], ... };
+  }
+
+  recordTimingData(idleDetectionMs: number, cycleDurationMs: number): void { ... }
+  getAdaptiveCompletionConfirmMs(): number { ... }
+  getTimingHistory(): TimingHistory { ... }
+  reset(): void { ... }
+}
+```
+
+### Step 3c: Extract `RespawnCycleMetrics` (~200 LOC)
+
+**Standalone metrics tracker**.
+
+Move properties:
+- `currentCycleMetrics`
+- `recentCycleMetrics[]`
+- `aggregateMetrics`
+- `MAX_CYCLE_METRICS_IN_MEMORY`
+
+Move methods:
+- `startCycleMetrics(idleReason)`
+- `recordCycleStep(step)`
+- `completeCycleMetrics(outcome, errorMessage?)`
+- `updateAggregateMetrics(metrics)`
+- `getAggregateMetrics()`
+- `getRecentCycleMetrics(limit?)`
+
+```typescript
+export class RespawnCycleMetricsTracker {
+  private currentCycleMetrics: Partial<RespawnCycleMetrics> | null = null;
+  private recentCycleMetrics: RespawnCycleMetrics[] = [];
+  private aggregateMetrics: RespawnAggregateMetrics;
+
+  startCycle(sessionId: string, cycleNumber: number, idleReason: string): void { ... }
+  recordStep(step: string): void { ... }
+  completeCycle(outcome: CycleOutcome, errorMessage?: string): RespawnCycleMetrics | null { ... }
+  getAggregate(): RespawnAggregateMetrics { ... }
+  getRecent(limit?: number): RespawnCycleMetrics[] { ... }
+  reset(): void { ... }
+}
+```
+
+**Callback**: `completeCycle()` returns the completed metrics so the controller can pass them to `adaptiveTiming.recordTimingData()`.
+
+### Step 3d: Extract `RespawnHealthCalculator` (~250 LOC)
+
+**Pure calculation — no state of its own**.
+
+Move methods:
+- `calculateHealthScore()`
+- `calculateCycleSuccessScore()`
+- `calculateCircuitBreakerScore()`
+- `calculateIterationProgressScore()`
+- `calculateAiCheckerScore()`
+- `calculateStuckRecoveryScore()`
+- `generateHealthRecommendations(components)`
+- `generateHealthSummary(score, status, components)`
+- `shouldSkipClear()` (belongs here since it's a pure calculation on token/config)
+
+```typescript
+export interface HealthInputs {
+  aggregateMetrics: RespawnAggregateMetrics;
+  circuitBreakerStatus: CircuitBreakerStatus;
+  iterationStallMetrics: { stallDurationMs: number; warningMs: number; criticalMs: number } | null;
+  aiCheckerState: { disabled: boolean; inCooldown: boolean; hasErrors: boolean };
+  stuckRecoveryCount: number;
+  maxStuckRecoveries: number;
+}
+
+export function calculateHealthScore(inputs: HealthInputs): RalphLoopHealthScore { ... }
+
+export function shouldSkipClear(
+  lastTokenCount: number,
+  skipClearThresholdPercent: number,
+  maxContextTokens: number
+): boolean { ... }
+```
+
+**Made as pure functions** (not a class) since they hold no state.
+
+### Step 3e: Keep in `respawn-controller.ts` (~2,200 LOC)
+
+The core state machine, idle detection, and AI checker integration stays:
+- State machine transitions (`setState`, `start`, `stop`, `pause`, `resume`)
+- Terminal data handling (`handleTerminalData`)
+- All 5 idle detection layers + hook signals
+- AI checker integration (`tryStartAiCheck`, `startAiCheck`, `startPlanCheck`)
+- Auto-accept logic
+- Step execution (`sendUpdateDocs`, `sendClear`, `sendInit`, `sendKickstart`)
+- Timer management (`startTrackedTimer`, `cancelTrackedTimer`)
+- Stuck-state detection and recovery
+- Action logging
+
+The class composes extracted modules:
+```typescript
+import { RespawnAdaptiveTiming } from './respawn-adaptive-timing.js';
+import { RespawnCycleMetricsTracker } from './respawn-metrics.js';
+import { calculateHealthScore, shouldSkipClear } from './respawn-health.js';
+import { isCompletionMessage, hasWorkingPattern, extractTokenCount } from './respawn-patterns.js';
+
+export class RespawnController extends EventEmitter {
+  private adaptiveTiming: RespawnAdaptiveTiming;
+  private cycleMetrics: RespawnCycleMetricsTracker;
+
+  calculateHealthScore(): RalphLoopHealthScore {
+    return calculateHealthScore({
+      aggregateMetrics: this.cycleMetrics.getAggregate(),
+      circuitBreakerStatus: this.session.ralphTracker.circuitBreakerStatus,
+      iterationStallMetrics: this.session.ralphTracker.getIterationStallMetrics(),
+      aiCheckerState: { ... },
+      stuckRecoveryCount: this.stuckRecoveryCount,
+      maxStuckRecoveries: this.config.maxStuckRecoveries ?? 3,
+    });
+  }
+}
+```
+
+---
+
+## 4. Split session.ts into focused modules
+
+**Current**: 2,418 lines, single `Session` class.
+**Risk**: LOW-MEDIUM — extractable pieces are utility-like with clear boundaries.
+
+### Coupling Analysis Summary
+
+| Module | Coupling | Extractability |
+|--------|----------|----------------|
+| CLI arg builder | NONE | HIGH — pure functions used at spawn time |
+| Auto-compact/clear | LOW | HIGH — self-contained automation with config |
+| Token tracking | LOW | MEDIUM — reads PTY output, writes state |
+| Task description cache | LOW | HIGH — separate LRU cache |
+| PTY + mux lifecycle | HIGH | KEEP — core of the class |
+| Tracker integration | HIGH | KEEP — event forwarding plumbing |
+
+### Target Structure
+
+```
+src/
+├── session.ts                    (~1,600 LOC — PTY lifecycle, terminal I/O,
+│                                   tracker integration, output processing,
+│                                   token tracking, state management)
+├── session-cli-builder.ts        (~250 LOC — Claude/OpenCode CLI arg construction)
+├── session-auto-ops.ts           (~300 LOC — auto-compact, auto-clear automation)
+└── session-task-cache.ts         (~100 LOC — task description LRU cache)
+```
+
+### Step 4a: Extract `SessionCliBuilder` (~250 LOC)
+
+**Pure functions — zero coupling to Session instance**.
+
+Move:
+- `buildClaudeArgs()` logic (currently inlined in `startInteractive` and `runPrompt`)
+- `buildOpenCodeArgs()` logic
+- Model mapping constants
+- Claude mode to flag mapping
+- Environment variable construction
+
+```typescript
+// src/session-cli-builder.ts
+export interface CliBuilderConfig {
+  claudeMode: ClaudeMode;
+  model?: string;
+  workingDir: string;
+  sessionId: string;
+  niceConfig?: NiceConfig;
+  isOpenCode?: boolean;
+  openCodeConfig?: OpenCodeConfig;
+}
+
+export function buildInteractiveArgs(config: CliBuilderConfig): string[] { ... }
+export function buildPromptArgs(config: CliBuilderConfig, prompt: string): string[] { ... }
+export function buildShellArgs(shell?: string): string[] { ... }
+export function buildClaudeEnv(config: CliBuilderConfig): Record<string, string> { ... }
+```
+
+### Step 4b: Extract `SessionAutoOps` (~300 LOC)
+
+**Self-contained automation with config-based thresholds**.
+
+Move properties:
+- `_autoCompactThreshold`
+- `_autoClearThreshold`
+- `_isAutoCompacting`
+- `_isAutoClearing`
+- `_autoCompactCount`
+- `_autoClearCount`
+- `_lastAutoCompactTime`
+- `_lastAutoClearTime`
+
+Move methods:
+- `checkAutoCompact(tokenCount)`
+- `performAutoCompact()`
+- `checkAutoClear(tokenCount)`
+- `performAutoClear()`
+- Auto-compact/clear threshold configuration
+
+```typescript
+export class SessionAutoOps extends EventEmitter {
+  constructor(
+    private writeCommand: (command: string) => Promise<void>,
+    private getTokenCount: () => number,
+    config: { compactThreshold: number; clearThreshold: number }
+  ) { ... }
+
+  /** Called after token count updates. Checks thresholds and triggers if needed. */
+  checkThresholds(tokenCount: number): void { ... }
+
+  updateConfig(config: { compactThreshold?: number; clearThreshold?: number }): void { ... }
+  getStats(): { autoCompactCount: number; autoClearCount: number; ... } { ... }
+}
+
+// Events: 'autoCompact', 'autoClear'
+```
+
+**In `Session`**: Compose and wire:
+```typescript
+private autoOps = new SessionAutoOps(
+  (cmd) => this.writeViaMux(cmd),
+  () => this._state.tokenCount,
+  { compactThreshold: 110_000, clearThreshold: 140_000 }
+);
+```
+
+### Step 4c: Extract `SessionTaskCache` (~100 LOC)
+
+**Isolated LRU cache for task descriptions**.
+
+Move:
+- `_taskDescriptionCache: LRUMap<number, { description: string; timestamp: number }>`
+- `_taskDescriptionMaxAge`
+- `findTaskDescriptionNear(lineNumber)`
+- `cacheTaskDescription(lineNumber, description)`
+
+```typescript
+export class SessionTaskCache {
+  private cache: LRUMap<number, { description: string; timestamp: number }>;
+  private maxAgeMs: number;
+
+  constructor(maxSize: number = 50, maxAgeMs: number = 30_000) { ... }
+
+  find(lineNumber: number, searchRadius: number = 50): string | null { ... }
+  add(lineNumber: number, description: string): void { ... }
+  clear(): void { ... }
+}
+```
+
+### Step 4d: Keep in `session.ts` (~1,600 LOC)
+
+The core stays together:
+- PTY process management (`spawn`, `kill`, `resize`, `writeViaMux`)
+- Data streaming pipeline (PTY → buffer → ANSI strip → JSON parse → events)
+- Tracker initialization and event forwarding (RalphTracker, BashToolParser, TaskTracker)
+- Output processing (message extraction, completion detection)
+- Token tracking (status line parsing)
+- State management (`toState()`, `updateState()`)
+- Session lifecycle (`startInteractive`, `startShell`, `runPrompt`)
+- CLI info detection (version, model, account)
+
+---
+
+## 5. Execution Order & Dependencies
+
+Execute in this order to minimize risk. Each step is independently deployable.
+
+```
+Step 1: types.ts split
+  ↓  (no runtime change, just file reorganization)
+Step 2a: RalphPlanTracker extraction
+  ↓  (independent of types split)
+Step 2b: RalphFixPlanWatcher extraction
+Step 2c: RalphStallDetector extraction
+Step 2d: RalphStatusParser extraction
+  ↓  (ralph-tracker.ts now ~1,800 LOC)
+Step 3a: RespawnPatterns extraction
+Step 3b: RespawnAdaptiveTiming extraction
+Step 3c: RespawnCycleMetrics extraction
+Step 3d: RespawnHealthCalculator extraction
+  ↓  (respawn-controller.ts now ~2,200 LOC)
+Step 4a: SessionCliBuilder extraction
+Step 4b: SessionAutoOps extraction
+Step 4c: SessionTaskCache extraction
+  ↓  (session.ts now ~1,600 LOC)
+```
+
+**Parallelization**: Steps 1, 2a-2d, 3a-3d, and 4a-4c can be done by separate agents in parallel since they touch different files. However, within each group, sequential execution is safer.
+
+### Risk Mitigation
+
+- **Barrel exports**: Every split uses delegation + barrel re-export so external consumers see zero API changes
+- **Event forwarding**: Sub-modules emit events, parent class forwards them — no event contract changes
+- **Incremental**: Each step can be verified independently with `tsc --noEmit` + `npm run lint`
+- **No test changes needed**: External API stays identical; existing tests continue to pass
+
+---
+
+## 6. Validation Checklist
+
+After each step, verify:
+
+- [ ] `tsc --noEmit` passes (no type errors)
+- [ ] `npm run lint` passes (no unused imports, etc.)
+- [ ] `npm run format:check` passes
+- [ ] `npx vitest run test/respawn-controller.test.ts` passes (for respawn splits)
+- [ ] `npx vitest run test/ralph-tracker.test.ts` passes (for ralph splits)
+- [ ] `npx vitest run test/session-manager.test.ts` passes (for session splits)
+- [ ] Dev server starts: `npx tsx src/index.ts web`
+- [ ] Existing sessions work (create, interact, delete)
+- [ ] Respawn cycle works (enable respawn, verify idle detection fires)
+- [ ] No new circular dependencies: `npx madge --circular src/`
+
+### Size Targets
+
+| File | Before | After |
+|------|--------|-------|
+| `src/types.ts` | 1,443 LOC | 1 LOC (re-export barrel) |
+| `src/ralph-tracker.ts` | 3,868 LOC | ~1,800 LOC |
+| `src/respawn-controller.ts` | 3,611 LOC | ~2,200 LOC |
+| `src/session.ts` | 2,418 LOC | ~1,600 LOC |
+| **Total new files** | — | 12 files |
+| **Net LOC change** | — | ~0 (refactor only) |

docs/phase1-implementation-plan.md

@@ -0,0 +1,738 @@
+# Phase 1 Implementation Plan: Quick Wins
+
+**Source**: `docs/code-structure-findings.md` (Phase 1 - Quick Wins section)
+**Estimated effort**: 1-2 days
+**Tasks**: 5 independent tasks (can be done in parallel unless noted)
+
+---
+
+## Safety Constraints
+
+Before starting ANY work, read and follow these rules:
+
+1. **Never run `npx vitest run`** (full suite) -- it kills tmux sessions. You are running inside a Codeman-managed tmux session.
+2. **Run individual tests only**: `npx vitest run test/<file>.test.ts`
+3. **Never test on port 3000** -- the live dev server runs there. Tests use ports 3150+.
+4. **After TypeScript changes**: Run `tsc --noEmit` to verify type checking passes.
+5. **Before considering done**: Run `npm run lint` and `npm run format:check` to ensure CI passes.
+6. **Never kill tmux sessions** -- check `echo $CODEMAN_MUX` first.
+
+---
+
+## Task Dependencies
+
+All 5 tasks are independent and can be done in parallel. However:
+- Task 1 (barrel exports) is a prerequisite if you want to update import sites to use the barrel after Task 3 (consolidate EXEC_TIMEOUT_MS). The EXEC_TIMEOUT_MS consolidation creates a new export that should be added to the barrel.
+- Task 2 (delete dead functions) removes functions that Task 1 would otherwise need to add to the barrel. Do Task 2 first or simultaneously with Task 1 to avoid adding exports for dead code.
+
+**Recommended order**: Task 2 -> Task 1 -> Task 3 -> Task 4 -> Task 5
+
+---
+
+## Task 1: Export Missing Functions from Utils Barrel
+
+**File**: `src/utils/index.ts`
+**Time**: ~30 minutes
+
+### Problem
+
+The barrel file (`src/utils/index.ts`) is missing exports for several functions that are defined in util modules, forcing consumers to use deep imports or preventing usage entirely.
+
+### Missing Exports
+
+From `src/utils/regex-patterns.ts`:
+- `createAnsiPatternFull()` -- factory for fresh ANSI regex (documented in CLAUDE.md)
+- `createAnsiPatternSimple()` -- factory for fresh ANSI regex (documented in CLAUDE.md)
+- `stripAnsi()` -- ANSI stripping utility
+- `SAFE_PATH_PATTERN` -- regex for safe file paths (currently deep-imported by `schemas.ts` and `tmux-manager.ts`)
+
+From `src/utils/token-validation.ts`:
+- `validateTokenCounts()` -- token count validation (documented in CLAUDE.md)
+- `validateTokensAndCost()` -- token + cost validation (documented in CLAUDE.md)
+
+**Note**: Do NOT export `isSimilar`, `isSimilarByDistance`, `levenshteinDistance`, or `normalizePhrase` from `string-similarity.ts` -- these are dead code (see Task 2).
+
+### Edit 1: Add missing regex-patterns exports
+
+**File**: `src/utils/index.ts`
+
+**Old code** (lines 13-18):
+```typescript
+export {
+  ANSI_ESCAPE_PATTERN_FULL,
+  ANSI_ESCAPE_PATTERN_SIMPLE,
+  TOKEN_PATTERN,
+  SPINNER_PATTERN,
+} from './regex-patterns.js';
+```
+
+**New code**:
+```typescript
+export {
+  ANSI_ESCAPE_PATTERN_FULL,
+  ANSI_ESCAPE_PATTERN_SIMPLE,
+  TOKEN_PATTERN,
+  SPINNER_PATTERN,
+  createAnsiPatternFull,
+  createAnsiPatternSimple,
+  stripAnsi,
+  SAFE_PATH_PATTERN,
+} from './regex-patterns.js';
+```
+
+### Edit 2: Add missing token-validation exports
+
+**File**: `src/utils/index.ts`
+
+**Old code** (line 19):
+```typescript
+export { MAX_SESSION_TOKENS } from './token-validation.js';
+```
+
+**New code**:
+```typescript
+export { MAX_SESSION_TOKENS, validateTokenCounts, validateTokensAndCost } from './token-validation.js';
+```
+
+### Optional follow-up: Update deep imports to use barrel
+
+These files currently deep-import `SAFE_PATH_PATTERN` and could be updated to use the barrel instead:
+
+- `src/web/schemas.ts` line 11: `import { SAFE_PATH_PATTERN } from '../utils/regex-patterns.js';` could become `import { SAFE_PATH_PATTERN } from '../utils/index.js';`
+- `src/tmux-manager.ts` line 44: `import { SAFE_PATH_PATTERN } from './utils/regex-patterns.js';` could become part of existing barrel import
+
+This is a low-priority cosmetic change. The barrel export itself is the important fix.
+
+### Verification
+
+```bash
+tsc --noEmit
+npm run lint
+```
+
+---
+
+## Task 2: Delete Dead Utility Functions
+
+**File**: `src/utils/string-similarity.ts`
+**Time**: ~15 minutes
+
+### Problem
+
+Four exported functions in `string-similarity.ts` are never imported anywhere in the codebase:
+- `levenshteinDistance()` (lines 27-69)
+- `isSimilar()` (lines 106-108)
+- `isSimilarByDistance()` (lines 123-125)
+- `normalizePhrase()` (lines 139-144)
+
+Only three functions are actually used (all by `ralph-tracker.ts` via the barrel):
+- `stringSimilarity()` -- uses `levenshteinDistance()` internally
+- `fuzzyPhraseMatch()` -- uses `normalizePhrase()` and `isSimilarByDistance()` internally
+- `todoContentHash()`
+
+### Strategy
+
+`levenshteinDistance()` is called by `stringSimilarity()`, and `normalizePhrase()` and `isSimilarByDistance()` are called by `fuzzyPhraseMatch()`. So they cannot be deleted -- they just need to be un-exported (made private to the module).
+
+`isSimilar()` is truly dead -- not called by anything. Delete it entirely.
+
+### Edit 1: Remove `export` from `levenshteinDistance`
+
+**File**: `src/utils/string-similarity.ts`
+
+**Old code** (line 27):
+```typescript
+export function levenshteinDistance(a: string, b: string): number {
+```
+
+**New code**:
+```typescript
+function levenshteinDistance(a: string, b: string): number {
+```
+
+### Edit 2: Delete `isSimilar` function entirely
+
+**File**: `src/utils/string-similarity.ts`
+
+**Old code** (lines 94-108):
+```typescript
+/**
+ * Check if two strings are similar within a given threshold.
+ *
+ * @param a - First string
+ * @param b - Second string
+ * @param threshold - Minimum similarity ratio (default: 0.85 = 85% similar)
+ * @returns True if similarity >= threshold
+ *
+ * @example
+ * isSimilar('COMPLETE', 'COMPLET', 0.85)  // true (87.5% similar)
+ * isSimilar('COMPLETE', 'DONE', 0.85)     // false (0% similar)
+ */
+export function isSimilar(a: string, b: string, threshold = 0.85): boolean {
+  return stringSimilarity(a, b) >= threshold;
+}
+```
+
+**New code**: (delete entirely -- replace with empty string)
+
+### Edit 3: Remove `export` from `isSimilarByDistance`
+
+**File**: `src/utils/string-similarity.ts`
+
+**Old code** (line 123):
+```typescript
+export function isSimilarByDistance(a: string, b: string, maxDistance = 2): boolean {
+```
+
+**New code**:
+```typescript
+function isSimilarByDistance(a: string, b: string, maxDistance = 2): boolean {
+```
+
+### Edit 4: Remove `export` from `normalizePhrase`
+
+**File**: `src/utils/string-similarity.ts`
+
+**Old code** (line 139):
+```typescript
+export function normalizePhrase(phrase: string): string {
+```
+
+**New code**:
+```typescript
+function normalizePhrase(phrase: string): string {
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+npx vitest run test/string-utilities.test.ts
+npm run lint
+```
+
+Note: If `test/string-utilities.test.ts` imports any of the now-unexported functions, those test imports will fail. Check the test file and remove tests for `isSimilar` (deleted) and update any direct tests for `levenshteinDistance`, `isSimilarByDistance`, `normalizePhrase` to test them indirectly through the public API (`stringSimilarity`, `fuzzyPhraseMatch`), or remove those tests.
+
+---
+
+## Task 3: Consolidate Duplicated `EXEC_TIMEOUT_MS` Constant
+
+**Files**:
+- `src/utils/claude-cli-resolver.ts` (line 17)
+- `src/utils/opencode-cli-resolver.ts` (line 16)
+- `src/tmux-manager.ts` (line 63) -- also has its own copy
+
+**Time**: ~15 minutes
+
+### Problem
+
+`EXEC_TIMEOUT_MS = 5000` is defined identically in three files. Changes need to happen in all three places.
+
+### Strategy
+
+Create a shared constant and export it. The natural home is a new config file since the existing config files (`buffer-limits.ts`, `map-limits.ts`) follow this pattern. However, to keep it minimal, we can add it to an existing config file or create a small one.
+
+**Recommended approach**: Add to `src/config/timing-config.ts` (new file) as a single constant. This file can grow later in Phase 6 to hold other timing constants.
+
+Alternatively, the simplest approach: export from one of the existing utils and import in the others. Since both CLI resolvers are in `src/utils/`, the cleanest approach is to put it in a shared location.
+
+### Option A: Add to existing config (simpler)
+
+Create `src/config/exec-timeout.ts`:
+
+**New file**: `src/config/exec-timeout.ts`
+```typescript
+/**
+ * Timeout for child process exec commands (e.g., `which claude`, `which opencode`, tmux commands).
+ * Used across CLI resolvers and tmux manager.
+ */
+export const EXEC_TIMEOUT_MS = 5000;
+```
+
+### Edit 1: Update `claude-cli-resolver.ts`
+
+**File**: `src/utils/claude-cli-resolver.ts`
+
+**Old code** (lines 11-17):
+```typescript
+import { execSync } from 'node:child_process';
+import { existsSync } from 'node:fs';
+import { delimiter, dirname, join } from 'node:path';
+import { homedir } from 'node:os';
+
+/** Timeout for exec commands (5 seconds) */
+const EXEC_TIMEOUT_MS = 5000;
+```
+
+**New code**:
+```typescript
+import { execSync } from 'node:child_process';
+import { existsSync } from 'node:fs';
+import { delimiter, dirname, join } from 'node:path';
+import { homedir } from 'node:os';
+import { EXEC_TIMEOUT_MS } from '../config/exec-timeout.js';
+```
+
+### Edit 2: Update `opencode-cli-resolver.ts`
+
+**File**: `src/utils/opencode-cli-resolver.ts`
+
+**Old code** (lines 10-16):
+```typescript
+import { execSync } from 'node:child_process';
+import { existsSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { homedir } from 'node:os';
+
+/** Timeout for exec commands (5 seconds) */
+const EXEC_TIMEOUT_MS = 5000;
+```
+
+**New code**:
+```typescript
+import { execSync } from 'node:child_process';
+import { existsSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { homedir } from 'node:os';
+import { EXEC_TIMEOUT_MS } from '../config/exec-timeout.js';
+```
+
+### Edit 3: Update `tmux-manager.ts`
+
+**File**: `src/tmux-manager.ts`
+
+**Old code** (line 63):
+```typescript
+const EXEC_TIMEOUT_MS = 5000;
+```
+
+**New code**:
+```typescript
+import { EXEC_TIMEOUT_MS } from './config/exec-timeout.js';
+```
+
+Note: `tmux-manager.ts` already has many imports at the top of the file. Add this import near the other local imports (around lines 43-56). The `const EXEC_TIMEOUT_MS = 5000;` on line 63 should be deleted entirely (replaced with the import).
+
+### Verification
+
+```bash
+tsc --noEmit
+npm run lint
+```
+
+---
+
+## Task 4: Add `z.infer` to Zod Schemas
+
+**Files**:
+- `src/web/schemas.ts` (add type exports)
+- `src/types.ts` (replace manual interfaces with `z.infer` re-exports where applicable)
+
+**Time**: ~2 hours
+
+### Problem
+
+All 30+ Zod schemas in `schemas.ts` define validation rules, but zero use `z.infer` to derive TypeScript types. Instead, `types.ts` manually duplicates interfaces that match the schemas. When a schema changes, the type must be manually updated too.
+
+### Strategy
+
+Add `z.infer` type exports to `schemas.ts` for each exported schema. This creates derived types as the single source of truth. For schemas that have corresponding manual interfaces in `types.ts`, the manual interface can be replaced with a re-export of the inferred type.
+
+**Important**: Not all schemas have matching interfaces in `types.ts`. The `RespawnConfig` interface in `types.ts` (line 395) has all required fields, while `RespawnConfigSchema` has all optional fields (it's for partial updates). These are NOT the same type and should NOT be unified.
+
+### Edit 1: Add inferred type exports to `schemas.ts`
+
+**File**: `src/web/schemas.ts`
+
+After each schema definition, add a corresponding type export. Add the following lines at the **end of the file** (after line 509):
+
+**Old code** (end of file, lines 506-509):
+```typescript
+    .optional(),
+});
+```
+
+Wait -- the end of file is actually at line 509 after the `RalphLoopStartSchema`. Add the type exports after the last schema:
+
+**Append to end of file** `src/web/schemas.ts`:
+
+```typescript
+
+// ========== Inferred Types ==========
+// Derive TypeScript types from Zod schemas (single source of truth)
+
+export type CreateSessionInput = z.infer<typeof CreateSessionSchema>;
+export type RunPromptInput = z.infer<typeof RunPromptSchema>;
+export type ResizeInput = z.infer<typeof ResizeSchema>;
+export type CreateCaseInput = z.infer<typeof CreateCaseSchema>;
+export type QuickStartInput = z.infer<typeof QuickStartSchema>;
+export type HookEventInput = z.infer<typeof HookEventSchema>;
+export type RespawnConfigInput = z.infer<typeof RespawnConfigSchema>;
+export type ConfigUpdateInput = z.infer<typeof ConfigUpdateSchema>;
+export type SettingsUpdateInput = z.infer<typeof SettingsUpdateSchema>;
+export type SessionInputWithLimitInput = z.infer<typeof SessionInputWithLimitSchema>;
+export type SessionNameInput = z.infer<typeof SessionNameSchema>;
+export type SessionColorInput = z.infer<typeof SessionColorSchema>;
+export type RalphConfigInput = z.infer<typeof RalphConfigSchema>;
+export type FixPlanImportInput = z.infer<typeof FixPlanImportSchema>;
+export type RalphPromptWriteInput = z.infer<typeof RalphPromptWriteSchema>;
+export type AutoClearInput = z.infer<typeof AutoClearSchema>;
+export type AutoCompactInput = z.infer<typeof AutoCompactSchema>;
+export type ImageWatcherInput = z.infer<typeof ImageWatcherSchema>;
+export type FlickerFilterInput = z.infer<typeof FlickerFilterSchema>;
+export type QuickRunInput = z.infer<typeof QuickRunSchema>;
+export type ScheduledRunInput = z.infer<typeof ScheduledRunSchema>;
+export type LinkCaseInput = z.infer<typeof LinkCaseSchema>;
+export type GeneratePlanInput = z.infer<typeof GeneratePlanSchema>;
+export type GeneratePlanDetailedInput = z.infer<typeof GeneratePlanDetailedSchema>;
+export type CancelPlanInput = z.infer<typeof CancelPlanSchema>;
+export type PlanTaskUpdateInput = z.infer<typeof PlanTaskUpdateSchema>;
+export type PlanTaskAddInput = z.infer<typeof PlanTaskAddSchema>;
+export type CpuLimitInput = z.infer<typeof CpuLimitSchema>;
+export type SubagentWindowStatesInput = z.infer<typeof SubagentWindowStatesSchema>;
+export type SubagentParentMapInput = z.infer<typeof SubagentParentMapSchema>;
+export type InteractiveRespawnInput = z.infer<typeof InteractiveRespawnSchema>;
+export type RespawnEnableInput = z.infer<typeof RespawnEnableSchema>;
+export type PushSubscribeInput = z.infer<typeof PushSubscribeSchema>;
+export type PushPreferencesUpdateInput = z.infer<typeof PushPreferencesUpdateSchema>;
+export type RalphLoopStartInput = z.infer<typeof RalphLoopStartSchema>;
+```
+
+### What NOT to do
+
+Do NOT replace the `RespawnConfig` interface in `types.ts` with `z.infer<typeof RespawnConfigSchema>`. The schema has all optional fields (for partial config updates), but the interface has required fields (for the full config object). These are intentionally different shapes.
+
+Similarly, do NOT try to unify every interface in `types.ts` with a schema -- most interfaces in `types.ts` represent internal domain objects (SessionState, TaskState, etc.) that have no corresponding Zod schema. The schemas only exist for API request validation.
+
+### Future opportunity
+
+In a future phase, route handlers in `server.ts` can use these inferred types for request body typing:
+```typescript
+const body = CreateSessionSchema.parse(request.body) as CreateSessionInput;
+```
+This task only adds the type exports. Migrating route handlers to use them is out of scope.
+
+### Verification
+
+```bash
+tsc --noEmit
+npm run lint
+npm run format:check
+```
+
+---
+
+## Task 5: Fix Weak `not.toThrow()` Tests with Behavioral Assertions
+
+**Files**:
+- `test/task-tracker.test.ts` -- 6 instances
+- `test/image-watcher.test.ts` -- 1 instance
+- `test/task-queue.test.ts` -- 1 instance
+- `test/hooks-config.test.ts` -- 1 instance
+- `test/session-manager.test.ts` -- 1 instance
+
+**Time**: ~1 hour
+
+### Problem
+
+10 tests only assert `not.toThrow()` without verifying the actual defensive behavior. These tests prove the code doesn't crash but don't verify it does the right thing.
+
+### Fix Strategy
+
+After each `not.toThrow()`, add a behavioral assertion that verifies the state is correct (e.g., no tasks were created, no side effects occurred).
+
+### Edit 1: `task-tracker.test.ts` -- null message (line 566)
+
+**File**: `test/task-tracker.test.ts`
+
+**Old code**:
+```typescript
+    it('should handle null message', () => {
+      expect(() => tracker.processMessage(null)).not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should handle null message', () => {
+      expect(() => tracker.processMessage(null)).not.toThrow();
+      expect(tracker.getAllTasks().size).toBe(0);
+      expect(tracker.getRunningCount()).toBe(0);
+    });
+```
+
+### Edit 2: `task-tracker.test.ts` -- message without content (line 569-571)
+
+**File**: `test/task-tracker.test.ts`
+
+**Old code**:
+```typescript
+    it('should handle message without content', () => {
+      expect(() => tracker.processMessage({ message: {} })).not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should handle message without content', () => {
+      expect(() => tracker.processMessage({ message: {} })).not.toThrow();
+      expect(tracker.getAllTasks().size).toBe(0);
+    });
+```
+
+### Edit 3: `task-tracker.test.ts` -- empty content array (line 573-575)
+
+**File**: `test/task-tracker.test.ts`
+
+**Old code**:
+```typescript
+    it('should handle empty content array', () => {
+      expect(() => tracker.processMessage({ message: { content: [] } })).not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should handle empty content array', () => {
+      expect(() => tracker.processMessage({ message: { content: [] } })).not.toThrow();
+      expect(tracker.getAllTasks().size).toBe(0);
+    });
+```
+
+### Edit 4: `task-tracker.test.ts` -- tool_result for unknown task (lines 577-590)
+
+**File**: `test/task-tracker.test.ts`
+
+**Old code**:
+```typescript
+    it('should handle tool_result for unknown task', () => {
+      expect(() => {
+        tracker.processMessage({
+          message: {
+            content: [{
+              type: 'tool_result',
+              tool_use_id: 'unknown-task',
+              is_error: false,
+              content: 'Done',
+            }],
+          },
+        });
+      }).not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should handle tool_result for unknown task', () => {
+      expect(() => {
+        tracker.processMessage({
+          message: {
+            content: [{
+              type: 'tool_result',
+              tool_use_id: 'unknown-task',
+              is_error: false,
+              content: 'Done',
+            }],
+          },
+        });
+      }).not.toThrow();
+      expect(tracker.getTask('unknown-task')).toBeUndefined();
+      expect(tracker.getAllTasks().size).toBe(0);
+    });
+```
+
+### Edit 5: `task-tracker.test.ts` -- empty terminal output (lines 592-595)
+
+**File**: `test/task-tracker.test.ts`
+
+**Old code**:
+```typescript
+    it('should handle empty terminal output', () => {
+      expect(() => tracker.processTerminalOutput('')).not.toThrow();
+      expect(() => tracker.processTerminalOutput('   ')).not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should handle empty terminal output', () => {
+      expect(() => tracker.processTerminalOutput('')).not.toThrow();
+      expect(() => tracker.processTerminalOutput('   ')).not.toThrow();
+      expect(tracker.getAllTasks().size).toBe(0);
+      expect(tracker.getRunningCount()).toBe(0);
+    });
+```
+
+### Edit 6: `image-watcher.test.ts` -- unwatchSession for non-watched session (line 123)
+
+**File**: `test/image-watcher.test.ts`
+
+**Old code**:
+```typescript
+    it('should be safe to call for non-watched session', () => {
+      expect(() => watcher.unwatchSession('nonexistent')).not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should be safe to call for non-watched session', () => {
+      expect(() => watcher.unwatchSession('nonexistent')).not.toThrow();
+      expect(watcher.getWatchedSessions()).toHaveLength(0);
+    });
+```
+
+### Edit 7: `task-queue.test.ts` -- dependencies on non-existent tasks (lines 538-542)
+
+**File**: `test/task-queue.test.ts`
+
+**Old code**:
+```typescript
+  it('should allow dependencies on non-existent tasks (just unsatisfied, not a cycle)', () => {
+    // Dependencies on non-existent tasks are valid - they just won't be satisfied
+    expect(() => {
+      queue.addTask({ prompt: 'Task D', dependencies: ['non-existent-id'] });
+    }).not.toThrow();
+  });
+```
+
+**New code**:
+```typescript
+  it('should allow dependencies on non-existent tasks (just unsatisfied, not a cycle)', () => {
+    // Dependencies on non-existent tasks are valid - they just won't be satisfied
+    let task: ReturnType<typeof queue.addTask> | undefined;
+    expect(() => {
+      task = queue.addTask({ prompt: 'Task D', dependencies: ['non-existent-id'] });
+    }).not.toThrow();
+    expect(task).toBeDefined();
+    expect(task!.dependencies).toEqual(['non-existent-id']);
+    // Task should be pending but blocked (dependency unsatisfied)
+    expect(queue.next()?.prompt).toBeUndefined();
+  });
+```
+
+Wait -- `queue.next()` returns `null` when no next task is available (all blocked). Let me adjust:
+
+**New code** (corrected):
+```typescript
+  it('should allow dependencies on non-existent tasks (just unsatisfied, not a cycle)', () => {
+    // Dependencies on non-existent tasks are valid - they just won't be satisfied
+    let task: ReturnType<typeof queue.addTask> | undefined;
+    expect(() => {
+      task = queue.addTask({ prompt: 'Task D', dependencies: ['non-existent-id'] });
+    }).not.toThrow();
+    expect(task).toBeDefined();
+    expect(task!.dependencies).toEqual(['non-existent-id']);
+    // Task exists but is blocked (dependency unsatisfied), so next() skips it
+    expect(queue.getAllTasks()).toHaveLength(1);
+    expect(queue.next()).toBeNull();
+  });
+```
+
+### Edit 8: `hooks-config.test.ts` -- valid JSON check (line 129)
+
+**File**: `test/hooks-config.test.ts`
+
+**Old code**:
+```typescript
+  it('should write valid JSON', () => {
+    writeHooksConfig(testDir);
+    const settingsPath = join(testDir, '.claude', 'settings.local.json');
+    const content = readFileSync(settingsPath, 'utf-8');
+    expect(() => JSON.parse(content)).not.toThrow();
+  });
+```
+
+**New code**:
+```typescript
+  it('should write valid JSON', () => {
+    writeHooksConfig(testDir);
+    const settingsPath = join(testDir, '.claude', 'settings.local.json');
+    const content = readFileSync(settingsPath, 'utf-8');
+    const parsed = JSON.parse(content);
+    expect(parsed).toBeDefined();
+    expect(typeof parsed).toBe('object');
+    expect(parsed.hooks).toBeDefined();
+  });
+```
+
+### Edit 9: `session-manager.test.ts` -- stopSession for non-existent (line 216)
+
+**File**: `test/session-manager.test.ts`
+
+**Old code**:
+```typescript
+    it('should handle non-existent session gracefully', async () => {
+      await expect(manager.stopSession('non-existent')).resolves.not.toThrow();
+    });
+```
+
+**New code**:
+```typescript
+    it('should handle non-existent session gracefully', async () => {
+      await expect(manager.stopSession('non-existent')).resolves.not.toThrow();
+      expect(manager.getSessionCount()).toBe(0);
+    });
+```
+
+### Verification
+
+Run each test file individually:
+
+```bash
+npx vitest run test/task-tracker.test.ts
+npx vitest run test/image-watcher.test.ts
+npx vitest run test/task-queue.test.ts
+npx vitest run test/hooks-config.test.ts
+npx vitest run test/session-manager.test.ts
+```
+
+**Important**: `hooks-config.test.ts` and `session-manager.test.ts` spawn real servers on ports 3130-3131. Only run them if you are NOT running other tests that use those ports.
+
+---
+
+## Final Verification Checklist
+
+After all 5 tasks are complete, run the following in order:
+
+```bash
+# 1. TypeScript type checking
+tsc --noEmit
+
+# 2. Linting
+npm run lint
+
+# 3. Formatting
+npm run format:check
+
+# 4. Run affected test files individually (NOT the full suite)
+npx vitest run test/string-utilities.test.ts
+npx vitest run test/task-tracker.test.ts
+npx vitest run test/image-watcher.test.ts
+npx vitest run test/task-queue.test.ts
+npx vitest run test/session-manager.test.ts
+npx vitest run test/hooks-config.test.ts
+```
+
+If any formatting issues arise, fix with:
+```bash
+npm run format
+```
+
+If any lint issues arise, fix with:
+```bash
+npm run lint:fix
+```
+
+### Summary of Changes
+
+| Task | Files Modified | Files Created |
+|------|---------------|---------------|
+| 1. Barrel exports | `src/utils/index.ts` | -- |
+| 2. Dead functions | `src/utils/string-similarity.ts` | -- |
+| 3. EXEC_TIMEOUT_MS | `src/utils/claude-cli-resolver.ts`, `src/utils/opencode-cli-resolver.ts`, `src/tmux-manager.ts` | `src/config/exec-timeout.ts` |
+| 4. z.infer types | `src/web/schemas.ts` | -- |
+| 5. Weak tests | `test/task-tracker.test.ts`, `test/image-watcher.test.ts`, `test/task-queue.test.ts`, `test/hooks-config.test.ts`, `test/session-manager.test.ts` | -- |
+
+**Total files modified**: 10
+**Total files created**: 1

docs/phase6-config-consolidation-plan.md

@@ -0,0 +1,689 @@
+# Phase 6 Implementation Plan: Config Consolidation
+
+**Source**: `docs/code-structure-findings.md` (Phase 6 — Config Consolidation)
+**Estimated effort**: 1 day
+**Tasks**: 8 tasks with dependencies (see dependency graph below)
+
+---
+
+## Safety Constraints
+
+Before starting ANY work, read and follow these rules:
+
+1. **Never run `npx vitest run`** (full suite) — it kills tmux sessions. You are running inside a Codeman-managed tmux session.
+2. **Run individual tests only**: `npx vitest run test/<file>.test.ts`
+3. **Never test on port 3000** — the live dev server runs there. Tests use ports 3150+.
+4. **After TypeScript changes**: Run `tsc --noEmit` to verify type checking passes.
+5. **Before considering done**: Run `npm run lint` and `npm run format:check` to ensure CI passes.
+6. **Never kill tmux sessions** — check `echo $CODEMAN_MUX` first.
+7. **Verify the dev server starts**: After each task, run `npx tsx src/index.ts web --port 3099 &` on a non-production port, confirm `curl -s http://localhost:3099/api/status | jq .status` returns `"ok"`, then kill the background process.
+
+---
+
+## Goal
+
+Consolidate ~70 scattered numeric constants from 15+ source files into 6 new domain-focused config files, eliminating cross-file duplicates (including a 5x-duplicated AI model string) and making all tuning knobs discoverable in `src/config/`.
+
+**Non-goal**: Moving every constant. Module-internal implementation details (like regex patterns, algorithm-specific magic numbers, or constants only used once in deeply coupled logic) stay where they are. The goal is discoverability of operational tuning knobs, not mechanical relocation.
+
+---
+
+## Design Decisions
+
+### What gets centralized (and why)
+
+Constants are candidates for centralization when they meet **any** of these criteria:
+
+1. **Duplicated across files** — DRY violation (e.g., `STATS_COLLECTION_INTERVAL_MS` in `server.ts` and `mux-routes.ts`, AI model string in 5 files)
+2. **Operational tuning knobs** — values an operator might want to adjust for performance, security, or behavior without understanding the implementation (e.g., SSE health check interval, auth session TTL, rate limits)
+3. **Cross-cutting concerns** — values that establish system-wide contracts (e.g., max terminal dimensions used by both server routes and frontend)
+
+### What stays in place (and why)
+
+Constants that are **internal implementation details** of a single module stay where they are:
+
+- **Algorithm parameters** — `TODO_SIMILARITY_THRESHOLD`, `adaptiveCompletionConfirmMs`, confidence weights. These are meaningless without understanding the algorithm.
+- **Display/UI formatting** — `TEXT_PREVIEW_LENGTH`, `SMART_TITLE_MAX_LENGTH`, `COMMAND_DISPLAY_LENGTH` in `subagent-watcher.ts`. Only used locally, tightly coupled to rendering logic.
+- **Module-internal timing** — `LINE_BUFFER_FLUSH_INTERVAL` in `session.ts`, `AI_CHECK_POLL_INTERVAL` in `ai-checker-base.ts`. Internal implementation of specific features.
+- **Frontend constants** — `constants.js` already centralizes frontend values well. Don't mix frontend and backend config.
+- **Respawn `DEFAULT_CONFIG`** — these are user-configurable defaults for the respawn config interface, not system constants. They live properly in `respawn-controller.ts`. The AI model/context defaults within it are replaced with imports from the new `ai-defaults.ts` (Task 5).
+- **Session auto-ops thresholds** — `AUTO_RETRY_DELAY_MS`, `COMPACT_COOLDOWN_MS`, etc. in `session-auto-ops.ts` are internal to that module's retry logic and already well-documented in place.
+
+### File organization: domain-based, not category-based
+
+A single `timing-config.ts` with 70 unrelated timing values would be worse than the current state — developers would need to grep it just like they grep the whole codebase now. Instead, constants are grouped by **the system they configure**:
+
+| New File | Domain | Developer Question It Answers |
+|----------|--------|-------------------------------|
+| `server-timing.ts` | Web server performance | "How do I tune SSE batching / terminal throughput?" |
+| `auth-config.ts` | Authentication & security | "What are the rate limits and session TTLs?" |
+| `tunnel-config.ts` | QR auth & Cloudflare tunnel | "What are the QR token rotation parameters?" |
+| `terminal-limits.ts` | Terminal dimensions & input | "What are the max cols/rows/input size?" |
+| `ai-defaults.ts` | AI checker model & context | "What model do the AI checkers use? What's the context limit?" |
+| `team-config.ts` | Agent Teams polling & caching | "How often does team polling run? What are the cache limits?" |
+
+---
+
+## Task Dependencies
+
+```
+Task 1 (server-timing.ts)
+Task 2 (auth-config.ts)
+Task 3 (tunnel-config.ts)
+Task 4 (terminal-limits.ts)
+Task 5 (ai-defaults.ts)
+Task 6 (team-config.ts)
+    └──> Task 7 (Fix remaining duplicates)
+         └──> Task 8 (Update CLAUDE.md + final verification)
+```
+
+**Tasks 1–6** are independent and can run in parallel.
+**Task 7** depends on Tasks 1–6 (needs the new config files to exist).
+**Task 8** depends on Task 7.
+
+---
+
+## Task 1: Create `src/config/server-timing.ts`
+
+**Estimated effort**: 30 minutes
+**Files created**: `src/config/server-timing.ts`
+**Files modified**: `src/web/server.ts`, `src/web/routes/mux-routes.ts`
+
+### Constants to extract from `src/web/server.ts`
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `TERMINAL_BATCH_INTERVAL` | `16` | Terminal data batching interval (60fps) |
+| `TASK_UPDATE_BATCH_INTERVAL` | `100` | Task event batching interval (ms) |
+| `STATE_UPDATE_DEBOUNCE_INTERVAL` | `500` | State persistence debounce (ms) |
+| `SESSIONS_LIST_CACHE_TTL` | `1000` | Sessions list cache TTL (ms) |
+| `SCHEDULED_CLEANUP_INTERVAL` | `300000` | Scheduled runs cleanup check (5 min) |
+| `SCHEDULED_RUN_MAX_AGE` | `3600000` | Completed scheduled run max age (1 hour) |
+| `SSE_HEALTH_CHECK_INTERVAL` | `30000` | SSE client health check (30s) |
+| `SESSION_LIMIT_WAIT_MS` | `5000` | Session limit retry wait (5s) |
+| `ITERATION_PAUSE_MS` | `2000` | Scheduled run iteration pause (2s) |
+| `BATCH_FLUSH_THRESHOLD` | `32768` | Terminal batch immediate flush threshold (32KB) |
+| `STATS_COLLECTION_INTERVAL_MS` | `2000` | Mux stats collection interval (2s) |
+
+### Implementation
+
+1. Create `src/config/server-timing.ts` with all 11 constants, preserving existing JSDoc comments.
+2. In `src/web/server.ts`: Remove the 11 local constant declarations (lines ~92–121). Add `import { TERMINAL_BATCH_INTERVAL, ... } from '../config/server-timing.js'`.
+3. In `src/web/routes/mux-routes.ts`: Remove the duplicate `STATS_COLLECTION_INTERVAL_MS` (line 10) and its comment. Add `import { STATS_COLLECTION_INTERVAL_MS } from '../../config/server-timing.js'`. This fixes a **duplicate constant** (finding #10).
+4. Run `tsc --noEmit`.
+
+### New file template
+
+```typescript
+/**
+ * @fileoverview Web server performance and scheduling constants.
+ *
+ * Controls terminal batching throughput, SSE health checking,
+ * state persistence debouncing, and scheduled run timing.
+ *
+ * @module config/server-timing
+ */
+
+// ============================================================================
+// Terminal & SSE Performance
+// ============================================================================
+
+/** Terminal data batching interval — targets 60fps (ms) */
+export const TERMINAL_BATCH_INTERVAL = 16;
+
+/** Immediate flush threshold for terminal batches (bytes).
+ * Set high (32KB) to allow effective batching; avg Ink events are ~14KB. */
+export const BATCH_FLUSH_THRESHOLD = 32 * 1024;
+
+/** Task event batching interval (ms) */
+export const TASK_UPDATE_BATCH_INTERVAL = 100;
+
+/** SSE client health check interval (ms) */
+export const SSE_HEALTH_CHECK_INTERVAL = 30 * 1000;
+
+// ============================================================================
+// State Persistence
+// ============================================================================
+
+/** State update debounce — batches expensive toDetailedState() calls (ms) */
+export const STATE_UPDATE_DEBOUNCE_INTERVAL = 500;
+
+/** Sessions list cache TTL — avoids re-serializing on every SSE init (ms) */
+export const SESSIONS_LIST_CACHE_TTL = 1000;
+
+// ============================================================================
+// Scheduled Runs
+// ============================================================================
+
+/** Scheduled runs cleanup check interval (ms) */
+export const SCHEDULED_CLEANUP_INTERVAL = 5 * 60 * 1000;
+
+/** Completed scheduled run max age before cleanup (ms) */
+export const SCHEDULED_RUN_MAX_AGE = 60 * 60 * 1000;
+
+/** Session limit retry wait before retrying (ms) */
+export const SESSION_LIMIT_WAIT_MS = 5000;
+
+/** Pause between scheduled run iterations (ms) */
+export const ITERATION_PAUSE_MS = 2000;
+
+// ============================================================================
+// Mux Stats
+// ============================================================================
+
+/** Mux stats collection interval (ms) */
+export const STATS_COLLECTION_INTERVAL_MS = 2000;
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+npx tsx src/index.ts web --port 3099 &
+curl -s http://localhost:3099/api/status | jq .status  # "ok"
+kill %1
+```
+
+---
+
+## Task 2: Create `src/config/auth-config.ts`
+
+**Estimated effort**: 20 minutes
+**Files created**: `src/config/auth-config.ts`
+**Files modified**: `src/web/middleware/auth.ts`, `src/hooks-config.ts`
+
+### Constants to extract from `src/web/middleware/auth.ts`
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `AUTH_SESSION_TTL_MS` | `86400000` | Auth session cookie TTL (24h) |
+| `MAX_AUTH_SESSIONS` | `100` | Max concurrent auth sessions |
+| `AUTH_FAILURE_MAX` | `10` | Max failed auth attempts per IP |
+| `AUTH_FAILURE_WINDOW_MS` | `900000` | Failed auth tracking window (15 min) |
+
+### Constants to extract from `src/hooks-config.ts`
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `HOOK_TIMEOUT_MS` | `10000` | Timeout for Claude Code hook commands |
+
+The `timeout: 10000` value is hardcoded 6 times in `hooks-config.ts` as inline literals. Extract to a single named constant.
+
+### Implementation
+
+1. Create `src/config/auth-config.ts` with the 5 constants.
+2. In `src/web/middleware/auth.ts`: Remove the 4 local constant declarations (lines 17–25). Add import from `../../config/auth-config.js`. Keep `AUTH_COOKIE_NAME` in place — it's a string identifier, not a tunable numeric constant.
+3. In `src/hooks-config.ts`: Replace all 6 inline `timeout: 10000` occurrences with `timeout: HOOK_TIMEOUT_MS`. Add import from `./config/auth-config.js`.
+4. Run `tsc --noEmit`.
+
+### New file template
+
+```typescript
+/**
+ * @fileoverview Authentication, rate limiting, and hook security constants.
+ *
+ * Controls auth session lifecycle, brute-force protection,
+ * and Claude Code hook timeouts.
+ *
+ * @module config/auth-config
+ */
+
+// ============================================================================
+// Session Cookies
+// ============================================================================
+
+/** Auth session cookie TTL — matches autonomous run length (ms) */
+export const AUTH_SESSION_TTL_MS = 24 * 60 * 60 * 1000;
+
+/** Max concurrent auth sessions per server */
+export const MAX_AUTH_SESSIONS = 100;
+
+// ============================================================================
+// Rate Limiting
+// ============================================================================
+
+/** Max failed auth attempts per IP before 429 rejection */
+export const AUTH_FAILURE_MAX = 10;
+
+/** Failed auth attempt tracking window (ms) */
+export const AUTH_FAILURE_WINDOW_MS = 15 * 60 * 1000;
+
+// ============================================================================
+// Hooks
+// ============================================================================
+
+/** Timeout for Claude Code hook curl commands (ms) */
+export const HOOK_TIMEOUT_MS = 10000;
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+npm run lint
+```
+
+---
+
+## Task 3: Create `src/config/tunnel-config.ts`
+
+**Estimated effort**: 20 minutes
+**Files created**: `src/config/tunnel-config.ts`
+**Files modified**: `src/tunnel-manager.ts`
+
+### Constants to extract from `src/tunnel-manager.ts`
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `QR_TOKEN_TTL_MS` | `60000` | QR token auto-rotation interval (60s) |
+| `QR_TOKEN_GRACE_MS` | `90000` | Grace period for previous token (90s) |
+| `SHORT_CODE_LENGTH` | `6` | Length of QR short code |
+| `QR_RATE_LIMIT_MAX` | `30` | Global QR attempt rate limit |
+| `QR_RATE_LIMIT_WINDOW_MS` | `60000` | QR rate limit reset window (60s) |
+| `URL_TIMEOUT_MS` | `30000` | Cloudflared URL fetch timeout (30s) |
+| `RESTART_DELAY_MS` | `5000` | Tunnel restart delay after crash (5s) |
+| `FORCE_KILL_MS` | `5000` | SIGTERM → SIGKILL escalation timeout (5s) |
+
+### Implementation
+
+1. Create `src/config/tunnel-config.ts` with all 8 constants.
+2. In `src/tunnel-manager.ts`: Remove the 8 local constant declarations (lines ~39–75). Add `import { QR_TOKEN_TTL_MS, ... } from './config/tunnel-config.js'`.
+3. Keep the `TUNNEL_URL_REGEX` in `tunnel-manager.ts` — it's a parsing detail, not a tuning knob.
+4. Run `tsc --noEmit`.
+
+### New file template
+
+```typescript
+/**
+ * @fileoverview Cloudflare tunnel and QR authentication constants.
+ *
+ * Controls QR token rotation timing, rate limiting,
+ * and tunnel process lifecycle.
+ *
+ * @module config/tunnel-config
+ */
+
+// ============================================================================
+// QR Token Rotation
+// ============================================================================
+
+/** QR token auto-rotation interval (ms) */
+export const QR_TOKEN_TTL_MS = 60_000;
+
+/** Grace period — previous token still valid during rotation (ms) */
+export const QR_TOKEN_GRACE_MS = 90_000;
+
+/** Length of the short code in QR URL path (chars) */
+export const SHORT_CODE_LENGTH = 6;
+
+// ============================================================================
+// QR Rate Limiting
+// ============================================================================
+
+/** Global rate limit for QR auth attempts across all IPs */
+export const QR_RATE_LIMIT_MAX = 30;
+
+/** QR rate limit reset window (ms) */
+export const QR_RATE_LIMIT_WINDOW_MS = 60_000;
+
+// ============================================================================
+// Tunnel Process Lifecycle
+// ============================================================================
+
+/** Max time to wait for cloudflared URL before timeout (ms) */
+export const URL_TIMEOUT_MS = 30_000;
+
+/** Restart delay after unexpected tunnel exit (ms) */
+export const RESTART_DELAY_MS = 5_000;
+
+/** SIGTERM → SIGKILL escalation timeout (ms) */
+export const FORCE_KILL_MS = 5_000;
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+```
+
+---
+
+## Task 4: Create `src/config/terminal-limits.ts`
+
+**Estimated effort**: 20 minutes
+**Files created**: `src/config/terminal-limits.ts`
+**Files modified**: `src/web/routes/session-routes.ts`
+
+### Constants to extract from `src/web/routes/session-routes.ts`
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `MAX_INPUT_LENGTH` | `65536` | Max input length per request (64KB) |
+| `MAX_TERMINAL_COLS` | `500` | Max terminal columns |
+| `MAX_TERMINAL_ROWS` | `200` | Max terminal rows |
+| `MAX_SESSION_NAME_LENGTH` | `128` | Max session name length (chars) |
+
+### Why a separate file instead of adding to `buffer-limits.ts`
+
+`buffer-limits.ts` covers memory buffer sizes (2MB terminal, 1MB text). These constants are **validation limits** for API inputs — different concern. A terminal resize request must not exceed `MAX_TERMINAL_COLS`; this has nothing to do with buffer trimming.
+
+### Implementation
+
+1. Create `src/config/terminal-limits.ts` with all 4 constants.
+2. In `src/web/routes/session-routes.ts`: Remove the 4 local constant declarations (lines 45–48). Add `import { MAX_INPUT_LENGTH, MAX_TERMINAL_COLS, MAX_TERMINAL_ROWS, MAX_SESSION_NAME_LENGTH } from '../../config/terminal-limits.js'`.
+3. Run `tsc --noEmit`.
+
+### New file template
+
+```typescript
+/**
+ * @fileoverview Terminal dimension and input validation limits.
+ *
+ * Used by API routes to validate resize, input, and session
+ * creation requests. Separate from buffer-limits.ts which
+ * controls memory buffer sizes.
+ *
+ * @module config/terminal-limits
+ */
+
+/** Max input length per API request (bytes) */
+export const MAX_INPUT_LENGTH = 64 * 1024;
+
+/** Max terminal columns for resize requests */
+export const MAX_TERMINAL_COLS = 500;
+
+/** Max terminal rows for resize requests */
+export const MAX_TERMINAL_ROWS = 200;
+
+/** Max session name length (chars) */
+export const MAX_SESSION_NAME_LENGTH = 128;
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+```
+
+---
+
+## Task 5: Create `src/config/ai-defaults.ts`
+
+**Estimated effort**: 30 minutes
+**Files created**: `src/config/ai-defaults.ts`
+**Files modified**: `src/respawn-controller.ts`, `src/ai-idle-checker.ts`, `src/ai-plan-checker.ts`, `src/web/routes/respawn-routes.ts`
+
+### Problem: AI model string duplicated 5 times
+
+The model identifier `'claude-opus-4-5-20251101'` appears in 5 places across 4 files. When the model changes, all 5 must be updated — a guaranteed source of bugs. The context limits (`16000`, `8000`) are similarly scattered across 3 files each.
+
+| Constant | Current Value | Duplicated In |
+|----------|---------------|---------------|
+| `AI_CHECK_MODEL` | `'claude-opus-4-5-20251101'` | `respawn-controller.ts` (×2: idle + plan), `ai-idle-checker.ts`, `ai-plan-checker.ts`, `respawn-routes.ts` (×2: idle + plan) |
+| `AI_IDLE_CHECK_MAX_CONTEXT` | `16000` | `respawn-controller.ts`, `ai-idle-checker.ts`, `respawn-routes.ts` |
+| `AI_PLAN_CHECK_MAX_CONTEXT` | `8000` | `respawn-controller.ts`, `ai-plan-checker.ts`, `respawn-routes.ts` |
+
+### Implementation
+
+1. Create `src/config/ai-defaults.ts` with the 3 constants.
+2. In `src/respawn-controller.ts` `DEFAULT_CONFIG` (line 538): Replace `aiIdleCheckModel: 'claude-opus-4-5-20251101'` with `aiIdleCheckModel: AI_CHECK_MODEL`, `aiIdleCheckMaxContext: 16000` with `aiIdleCheckMaxContext: AI_IDLE_CHECK_MAX_CONTEXT`, `aiPlanCheckModel: 'claude-opus-4-5-20251101'` with `aiPlanCheckModel: AI_CHECK_MODEL`, `aiPlanCheckMaxContext: 8000` with `aiPlanCheckMaxContext: AI_PLAN_CHECK_MAX_CONTEXT`. Add import from `./config/ai-defaults.js`.
+3. In `src/ai-idle-checker.ts` `DEFAULT_AI_CHECK_CONFIG` (line 46): Replace `model: 'claude-opus-4-5-20251101'` with `model: AI_CHECK_MODEL`, `maxContextChars: 16000` with `maxContextChars: AI_IDLE_CHECK_MAX_CONTEXT`. Add import from `./config/ai-defaults.js`.
+4. In `src/ai-plan-checker.ts` `DEFAULT_PLAN_CHECK_CONFIG` (line 45): Replace `model: 'claude-opus-4-5-20251101'` with `model: AI_CHECK_MODEL`, `maxContextChars: 8000` with `maxContextChars: AI_PLAN_CHECK_MAX_CONTEXT`. Add import from `./config/ai-defaults.js`.
+5. In `src/web/routes/respawn-routes.ts` config merge block (lines 173–179): Replace all 4 inline fallback values with imports from `../../config/ai-defaults.js`.
+6. Run `tsc --noEmit`.
+
+### New file template
+
+```typescript
+/**
+ * @fileoverview Default model and context limits for AI-powered checkers.
+ *
+ * Centralizes the AI model identifier and context window sizes used by
+ * the idle checker, plan checker, respawn controller defaults, and
+ * respawn route fallbacks. Change the model here when upgrading.
+ *
+ * @module config/ai-defaults
+ */
+
+/** Default model for AI idle and plan checkers */
+export const AI_CHECK_MODEL = 'claude-opus-4-5-20251101';
+
+/** Max context chars for idle checker (~4k tokens) */
+export const AI_IDLE_CHECK_MAX_CONTEXT = 16000;
+
+/** Max context chars for plan checker (~2k tokens, plan mode UI is compact) */
+export const AI_PLAN_CHECK_MAX_CONTEXT = 8000;
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+# Verify no remaining hardcoded model strings
+grep -rn 'claude-opus-4-5-20251101' src/  # Should only appear in config/ai-defaults.ts
+```
+
+---
+
+## Task 6: Create `src/config/team-config.ts`
+
+**Estimated effort**: 15 minutes
+**Files created**: `src/config/team-config.ts`
+**Files modified**: `src/team-watcher.ts`
+
+### Constants to extract from `src/team-watcher.ts`
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `TEAM_POLL_INTERVAL_MS` | `30000` | Team directory poll interval (30s) |
+| `MAX_CACHED_TEAMS` | `50` | LRU cache size for team configs |
+| `MAX_CACHED_TASKS` | `200` | LRU cache size for team tasks + inboxes |
+
+### Why centralize these
+
+Team polling frequency and cache sizes are operational knobs that affect both performance (polling too often wastes CPU) and responsiveness (polling too rarely means stale team state in the UI). They're also the kind of values a developer tuning for a large team deployment would want to find quickly. `MAX_CACHED_TASKS` is used for both the task cache and inbox cache — worth documenting.
+
+### Implementation
+
+1. Create `src/config/team-config.ts` with the 3 constants.
+2. In `src/team-watcher.ts`: Remove the 3 local constants (lines 23–25). Add `import { TEAM_POLL_INTERVAL_MS, MAX_CACHED_TEAMS, MAX_CACHED_TASKS } from './config/team-config.js'`. Note: rename `POLL_INTERVAL_MS` → `TEAM_POLL_INTERVAL_MS` to avoid ambiguity with the identically-named constant in `subagent-watcher.ts`.
+3. Update the usage site: `setInterval(... POLL_INTERVAL_MS)` → `setInterval(... TEAM_POLL_INTERVAL_MS)`.
+4. Run `tsc --noEmit`.
+
+### New file template
+
+```typescript
+/**
+ * @fileoverview Agent Teams polling and cache configuration.
+ *
+ * Controls how frequently TeamWatcher polls ~/.claude/teams/
+ * and how many teams/tasks are cached in memory.
+ *
+ * @module config/team-config
+ */
+
+/** Team directory poll interval (ms) */
+export const TEAM_POLL_INTERVAL_MS = 30_000;
+
+/** Max cached team configs (LRU eviction) */
+export const MAX_CACHED_TEAMS = 50;
+
+/** Max cached team tasks and inbox messages (LRU eviction).
+ * Used for both teamTasks and inboxCache maps. */
+export const MAX_CACHED_TASKS = 200;
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+```
+
+---
+
+## Task 7: Fix remaining cross-file duplicates
+
+**Estimated effort**: 30 minutes
+**Files modified**: `src/index.ts`, `src/subagent-watcher.ts`
+
+### Duplicate 1: `STATS_COLLECTION_INTERVAL_MS`
+
+Already fixed in Task 1 — both `server.ts` and `mux-routes.ts` now import from `server-timing.ts`.
+
+### Duplicate 2: AI model string
+
+Already fixed in Task 5 — all 5 occurrences now import from `ai-defaults.ts`.
+
+### Duplicate 3: `MAX_SCREENSHOT_SIZE` / `MAX_TEXT_FILE_SIZE` / `MAX_RAW_FILE_SIZE`
+
+These file size limits in `file-routes.ts` and `system-routes.ts` are **API-specific validation limits**. They're only used in their respective route files and aren't duplicated. **Leave in place** — they're local to their route module and well-commented.
+
+### Action A: Move `MAX_CONSECUTIVE_ERRORS` and `ERROR_RESET_MS` to config
+
+`src/index.ts` has two process-level constants that are operational tuning knobs:
+
+| Constant | Value | Purpose |
+|----------|-------|---------|
+| `MAX_CONSECUTIVE_ERRORS` | `5` | Max consecutive unhandled errors before process exit |
+| `ERROR_RESET_MS` | `60000` | Error counter reset interval (1 min) |
+
+These belong in a config file since they control server reliability behavior. Add them to `src/config/server-timing.ts` (they're server operational constants).
+
+1. Add to `src/config/server-timing.ts`:
+   ```typescript
+   // ============================================================================
+   // Process Error Recovery
+   // ============================================================================
+
+   /** Max consecutive unhandled errors before auto-restart */
+   export const MAX_CONSECUTIVE_ERRORS = 5;
+
+   /** Error counter reset interval — forgives errors after quiet period (ms) */
+   export const ERROR_RESET_MS = 60_000;
+   ```
+2. In `src/index.ts`: Remove lines 19–20, add import from `./config/server-timing.js`.
+3. Run `tsc --noEmit`.
+
+### Action B: Fix `MAX_TRACKED_AGENTS` shadow in `subagent-watcher.ts`
+
+`subagent-watcher.ts` defines its own `MAX_TRACKED_AGENTS = 500` locally instead of importing the identical value from `config/map-limits.ts`. This is a latent bug — if someone changes the config value, the subagent watcher's copy stays stale.
+
+1. In `src/subagent-watcher.ts`: Remove the local `MAX_TRACKED_AGENTS` constant. Add `import { MAX_TRACKED_AGENTS } from './config/map-limits.js'` (the value there is `MAX_TODOS_PER_SESSION = 500` — **verify** the map-limits constant is actually named `MAX_TRACKED_AGENTS` or if it needs to be added). If the constant doesn't exist in `map-limits.ts` under that name, add it.
+2. Run `tsc --noEmit`.
+
+### Verification
+
+```bash
+tsc --noEmit
+npm run lint
+npm run format:check
+```
+
+---
+
+## Task 8: Update CLAUDE.md and final verification
+
+**Estimated effort**: 20 minutes
+**Files modified**: `CLAUDE.md`
+
+### Updates to CLAUDE.md
+
+1. **Config Files table** (`src/config/`): Add the 6 new files:
+
+   | File | Purpose |
+   |------|---------|
+   | `buffer-limits.ts` | Terminal/text buffer size limits |
+   | `map-limits.ts` | Global limits for Maps, sessions, watchers |
+   | `exec-timeout.ts` | Execution timeout configuration |
+   | `server-timing.ts` | Web server batching, SSE, scheduled run timing |
+   | `auth-config.ts` | Auth session TTL, rate limits, hook timeout |
+   | `tunnel-config.ts` | QR token rotation, tunnel process lifecycle |
+   | `terminal-limits.ts` | Terminal dimension and input validation limits |
+   | `ai-defaults.ts` | AI checker model and context limits |
+   | `team-config.ts` | Agent Teams polling and cache sizes |
+
+2. **Import Conventions** section: Add:
+   ```
+   - **Config**: Import from specific files: `import { MAX_TERMINAL_COLS } from './config/terminal-limits'`
+   ```
+
+3. **Phase 6 status** in `docs/code-structure-findings.md`: Mark as COMPLETE with summary of what was done.
+
+### Final verification checklist
+
+```bash
+# Type checking
+tsc --noEmit
+
+# Linting
+npm run lint
+
+# Formatting
+npm run format:check
+
+# Dev server starts
+npx tsx src/index.ts web --port 3099 &
+curl -s http://localhost:3099/api/status | jq .status  # "ok"
+kill %1
+
+# Verify no remaining duplicates
+grep -rn 'STATS_COLLECTION_INTERVAL_MS' src/  # Should only appear in config + import sites
+grep -rn 'timeout: 10000' src/hooks-config.ts  # Should be 0 — all replaced with HOOK_TIMEOUT_MS
+grep -rn 'claude-opus-4-5-20251101' src/       # Should only appear in config/ai-defaults.ts
+```
+
+---
+
+## What is NOT in scope (and why)
+
+These constants were considered but deliberately left in their current files:
+
+### Respawn controller defaults (`src/respawn-controller.ts`)
+
+The `DEFAULT_CONFIG` object (lines 538–578) contains ~30 default values for the `RespawnConfig` interface. These are **user-facing configuration defaults**, not system constants — they're the starting values for a config object that users can modify via the API and UI. Centralizing them would break the locality between the config interface definition and its defaults. They already have excellent JSDoc with `@default` tags. The only values extracted are the AI model/context constants (Task 5) which are duplicated in other files.
+
+### Subagent watcher timing (`src/subagent-watcher.ts`)
+
+The 18 constants at lines 129–158 are all internal to the subagent watcher's polling/lifecycle algorithm. Moving them to a config file would force developers to context-switch between two files to understand the polling logic. They're already grouped with clear comments. Exception: `MAX_TRACKED_AGENTS` is consolidated with `map-limits.ts` (Task 7B) since it duplicates a global limit.
+
+### Session auto-ops timing (`src/session-auto-ops.ts`)
+
+The 8 constants at lines 19–40 are internal to the auto-compact/clear retry state machine. They form a coherent group that's meaningless without the surrounding implementation context.
+
+### Run summary constants (`src/run-summary.ts`)
+
+`MAX_EVENTS`, `TRIM_TO_EVENTS`, `TOKEN_MILESTONE_INTERVAL`, `STATE_STUCK_WARNING_MS`, `STATE_STUCK_CHECK_INTERVAL` — all module-internal. The buffer-style limits (`MAX_EVENTS`/`TRIM_TO_EVENTS`) follow the same pattern as `buffer-limits.ts` but are only used in this one file.
+
+### Frontend (`src/web/public/constants.js`)
+
+Already well-centralized. Frontend and backend run in different environments — mixing them in TypeScript config files would create import problems. If frontend constants need expansion, do it in `constants.js`. Note: `app.js` has 2 inline uses of `256 * 1024` that should use the existing `TERMINAL_TAIL_SIZE` from `constants.js` — a minor cleanup that can be done opportunistically but is not worth a task here.
+
+### Tmux manager timing (`src/tmux-manager.ts`)
+
+The 6 constants (lines 65–78) are internal to tmux process lifecycle management. They're low-level retry/wait values that are meaningless without understanding the tmux spawn sequence.
+
+### Process-internal constants
+
+`image-watcher.ts`, `bash-tool-parser.ts`, `transcript-watcher.ts`, `ralph-tracker.ts`, `task-tracker.ts`, `file-stream-manager.ts`, `session-lifecycle-log.ts`, `session-task-cache.ts`, `respawn-metrics.ts`, `respawn-adaptive-timing.ts`, `ai-checker-base.ts` — all have module-local constants that are internal implementation details.
+
+### `localhost:3000` default URL
+
+The string `'http://localhost:3000'` or port `3000` appears as a fallback default in ~5 files (`session-cli-builder.ts`, `tmux-manager.ts`, `tunnel-manager.ts`, `server.ts`, CLI). While technically duplicated, extracting it provides little value — each usage has a different fallback chain (env var → config → hardcoded) and the port is also baked into systemd service files and documentation. The risk of a missed update is low since port 3000 is deeply conventional.
+
+### `SAVE_DEBOUNCE_MS = 500` in `state-store.ts` / `push-store.ts`
+
+Same value (500ms), but they debounce different persistence targets (state.json vs push-subscriptions.json). If one needed faster/slower debouncing, they'd diverge. Coupling them would be misleading.
+
+---
+
+## Summary
+
+| Metric | Before | After |
+|--------|--------|-------|
+| Config files in `src/config/` | 3 | 9 |
+| Constants centralized | ~25 | ~65 |
+| Cross-file duplicates | 9+ (`STATS_COLLECTION_INTERVAL_MS`, `timeout: 10000` ×6, AI model ×5, context limits ×3 each, `MAX_TRACKED_AGENTS`) | 0 |
+| Files with `timeout: 10000` inline | 1 (6 occurrences) | 0 |
+| Files with hardcoded AI model string | 4 (5 occurrences) | 1 (config only) |
+| Files modified | — | 11 |
+| Files created | — | 6 |

docs/phase7-test-infrastructure-plan.md

@@ -0,0 +1,953 @@
+# Phase 7 Implementation Plan: Test Infrastructure
+
+**Source**: `docs/code-structure-findings.md` (Phase 7 — Test Infrastructure)
+**Estimated effort**: 2–3 days
+**Tasks**: 11 tasks with dependencies (see dependency graph below)
+
+---
+
+## Safety Constraints
+
+Before starting ANY work, read and follow these rules:
+
+1. **Never run `npx vitest run`** (full suite) — it kills tmux sessions. You are running inside a Codeman-managed tmux session.
+2. **Run individual tests only**: `npx vitest run test/<file>.test.ts`
+3. **Never test on port 3000** — the live dev server runs there. Tests use ports 3150+.
+4. **After TypeScript changes**: Run `tsc --noEmit` to verify type checking passes.
+5. **Before considering done**: Run `npm run lint` and `npm run format:check` to ensure CI passes.
+6. **Never kill tmux sessions** — check `echo $CODEMAN_MUX` first.
+7. **Port assignments for this phase**: New tests use ports 3220–3229 (see individual tasks for assignments).
+
+---
+
+## Goal
+
+Eliminate duplicated test mocks, activate the unused `respawn-test-utils.ts` utilities, and add route-level test coverage for the server's 12 route modules — the single largest untested area in the codebase (162 route handlers, 0 dedicated tests).
+
+**Non-goals**:
+- Full end-to-end integration tests (those require real Claude CLI / tmux sessions)
+- 100% route coverage in this phase — focus on the highest-value route modules first
+- Refactoring test patterns in existing passing tests that don't use shared mocks
+- Migrating `vi.mock()`-based module replacement mocks (different pattern, see Task 6/7)
+
+---
+
+## Current State
+
+### Mock Duplication (Finding #9)
+
+`MockSession` is defined **4 times** across test files with varying levels of completeness:
+
+| File | Properties | Methods | EventEmitter | Notes |
+|------|-----------|---------|-------------|-------|
+| `test/respawn-test-utils.ts` | 6 | 20+ | Yes | **Most complete**. Includes terminal simulation, token count, ANSI output, plan mode prompts. **Never imported by any test.** |
+| `test/respawn-controller.test.ts` | 6 | 9 | Yes | Subset of respawn-test-utils. Missing token simulation, ANSI helpers. |
+| `test/respawn-team-awareness.test.ts` | ~6 | ~9 | Yes | Near-copy of respawn-controller.test.ts version. |
+| `test/session-manager.test.ts` | 4 | 8 | Yes | **Inside `vi.mock()` factory** — replaces `../src/session.js` module. Different shape: `start()`/`stop()`/`toState()`/`sendInput()` for lifecycle testing. |
+
+`MockStateStore` is defined **2 times** (both inside `vi.mock()` factories):
+
+| File | Shape | Methods | Mock Pattern |
+|------|-------|---------|-------------|
+| `test/session-manager.test.ts` | `{ sessions, config }` | `getConfig`, `getSessions`, `getSession`, `setSession`, `removeSession` | `vi.mock('../src/state-store.js')` |
+| `test/ralph-loop.test.ts` | `{ ralphLoop, tasks, config }` | `getConfig`, `getRalphLoopState`, `setRalphLoopState`, `getTasks`, `setTask`, `removeTask` | `vi.mock('../src/state-store.js')` |
+
+### Important: Two distinct mocking patterns
+
+The codebase uses two different mocking patterns that require different migration strategies:
+
+1. **Direct instantiation** (respawn-controller, respawn-team-awareness): `MockSession` is defined at file scope and instantiated directly in tests. These can be migrated to shared mocks via simple import replacement.
+
+2. **Module replacement** (session-manager, ralph-loop): Mocks are defined inside `vi.mock()` factories that replace entire modules (`../src/session.js`, `../src/state-store.js`). These factories run in an isolated scope and return `{ Session: MockClass }` or `{ getStore: vi.fn(() => instance) }`. Migrating these requires either `vi.hoisted()` or restructuring the test's module mocking — higher risk for limited benefit.
+
+### Unused Test Utilities
+
+`test/respawn-test-utils.ts` exports these utilities that **no test file imports**:
+
+- `TimeController` / `createTimeController()` — abstraction over vitest fake timers
+- `MockAiIdleChecker` / `MockAiPlanChecker` — fully mocked AI checkers with result queueing
+- `createStateTracker()` / `createEventRecorder()` — state transition and event recording
+- `FAST_TEST_CONFIG` / `AI_ENABLED_TEST_CONFIG` — pre-configured RespawnConfig objects
+- `waitForState()` / `waitForEvent()` / `createDeferred()` — async test helpers
+- `terminalOutputs` — factory object for common terminal output patterns
+
+### Route Test Coverage
+
+Currently **zero** dedicated tests for the 12 route modules in `src/web/routes/`. The existing test files that touch API endpoints:
+
+| Test File | What It Tests | Approach |
+|-----------|--------------|----------|
+| `test/api-responses.test.ts` | Response structure validation | Imports types, no HTTP calls |
+| `test/api-generate-plan.test.ts` | Plan generation API | Mocks validation logic, Port 3191 declared |
+| `test/auth-security.test.ts` | Auth middleware | Integration tests with WebServer, Ports 3160/3161 |
+| `test/qr-auth.test.ts` | QR authentication | Integration + unit tests, Port 3162 |
+
+None of these test the route handlers themselves with real HTTP requests against a running Fastify instance.
+
+---
+
+## Design Decisions
+
+### Shared mocks: Superset strategy
+
+Rather than creating a lowest-common-denominator mock, `MockSession` in `test/mocks/` will be the **superset** from `respawn-test-utils.ts` (the most complete version). Test files that need a simpler mock can just ignore the extra methods — having unused methods costs nothing, but missing methods forces local re-definition.
+
+### vi.mock() tests: Don't migrate
+
+The `session-manager.test.ts` and `ralph-loop.test.ts` tests define mocks inside `vi.mock()` factories. These use **module-level replacement** (replacing `../src/session.js` and `../src/state-store.js` entirely), which is fundamentally different from the direct-instantiation pattern. Migrating them would require `vi.hoisted()` or factory restructuring — high complexity for limited benefit since these mocks are already working. We leave these as-is and create the shared mocks for **new** tests and for the two direct-instantiation tests (Tasks 4–5).
+
+### MockStateStore: Union of both shapes
+
+The shared `MockStateStore` in `test/mocks/` will include methods from both existing definitions (session management + Ralph loop), so any **new** test can use it. Methods default to no-ops via `vi.fn()`. Existing `vi.mock()`-based tests are not migrated.
+
+### Route testing strategy: Lightweight Fastify instances
+
+Each route test file will:
+1. Create a minimal `Fastify` instance
+2. Register **only** the route module under test
+3. Provide a mock context object satisfying the port interfaces
+4. Use `app.inject()` (Fastify's built-in test helper) — no real HTTP, no port needed
+
+This avoids port conflicts entirely and runs fast. Only tests that need SSE or WebSocket behavior will use a real listening server with assigned ports.
+
+### Port assignments (for tests needing real servers)
+
+| Port | Test File | Purpose |
+|------|-----------|---------|
+| 3220 | `test/routes/session-routes.test.ts` | SSE integration (if needed) |
+| 3221 | `test/routes/system-routes.test.ts` | Status/stats endpoints |
+| 3222 | `test/routes/respawn-routes.test.ts` | Respawn API |
+| 3223 | `test/routes/ralph-routes.test.ts` | Ralph API |
+| 3224–3229 | Reserved | Future route tests |
+
+Most tests should NOT need real ports — `app.inject()` is preferred. Verified: ports 3220–3229 are completely unused by existing tests (highest used port is 3211 in `opencode-resize.test.ts`).
+
+---
+
+## Task Dependencies
+
+```
+Task 1 (Consolidate MockSession)
+Task 2 (Consolidate MockStateStore)
+    └──> Task 3 (Create test/mocks/ barrel)
+         ├──> Task 4 (Migrate respawn-controller.test.ts)
+         ├──> Task 5 (Migrate respawn-team-awareness.test.ts)
+         └──> Task 6 (Route test scaffold + helpers)
+              ├──> Task 7 (Session routes tests)
+              └──> Task 8 (System + respawn routes tests)
+
+Task 9 (Slim down respawn-test-utils.ts) — depends on Tasks 4, 5
+```
+
+**Tasks 1–2** are independent and can run in parallel.
+**Task 3** depends on Tasks 1–2.
+**Tasks 4–6** depend on Task 3 and can run in parallel.
+**Tasks 7–8** depend on Task 6 and can run in parallel.
+**Task 9** depends on Tasks 4, 5 (must verify migrations work before removing duplicates from source).
+
+---
+
+## Task 1: Consolidate MockSession into `test/mocks/mock-session.ts`
+
+**Estimated effort**: 2 hours
+**Files created**: `test/mocks/mock-session.ts`
+**Files modified**: None yet (consumers migrate in Tasks 4–5)
+
+### Source
+
+The canonical MockSession comes from `test/respawn-test-utils.ts` (lines 89–241). It is the most complete version with:
+
+- All properties needed by `RespawnController`: `id`, `workingDir`, `status`, `writeBuffer`, `terminalBuffer`, `muxName`
+- `write()` / `writeViaMux()` for input simulation
+- Buffer inspection: `lastWrite`, `hasWritten(pattern)`, `clearWriteBuffer()`
+- Terminal simulation: `simulateTerminalOutput()`, `simulatePrompt()`, `simulateReady()`, `simulateCompletionMessage()`, `simulateWorking()`, `simulateClearComplete()`, `simulateInitComplete()`, `simulatePlanModePrompt()`, `simulateElicitationDialog()`, `simulateTokenCount()`, `simulateAnsiOutput()`
+- Lifecycle: `close()`
+
+### Implementation
+
+1. Create `test/mocks/` directory.
+2. Create `test/mocks/mock-session.ts`:
+   - Copy the `MockSession` class **exactly** from `test/respawn-test-utils.ts` (lines 89–241)
+   - Copy `terminalOutputs` helper object (tightly coupled to mock)
+   - Copy `createMockSession()` factory function
+   - Export all three: `export { MockSession, createMockSession, terminalOutputs }`
+   - Ensure all `vi` imports come from `vitest`
+
+**CRITICAL**: Copy the source verbatim — do NOT rewrite the simulation methods. The respawn controller's detection logic matches specific output patterns (e.g., `'\u276f '` for prompt, `'\u273b Worked for'` for completion). Using different patterns would cause test failures.
+
+### Template
+
+```typescript
+/**
+ * Shared MockSession for tests that need terminal simulation.
+ *
+ * Copied from test/respawn-test-utils.ts (the canonical, most complete version).
+ * Used by respawn, route, and subagent tests.
+ */
+import { EventEmitter } from 'node:events';
+
+// Copy MockSession class exactly from test/respawn-test-utils.ts lines 89–241
+export class MockSession extends EventEmitter {
+  // ... (copy verbatim from respawn-test-utils.ts)
+}
+
+/**
+ * Factory for common terminal output strings.
+ * Must match the patterns used in MockSession's simulate* methods.
+ */
+export const terminalOutputs = {
+  // ... (copy verbatim from respawn-test-utils.ts)
+};
+
+/**
+ * Convenience factory.
+ */
+export function createMockSession(id?: string): MockSession {
+  return new MockSession(id);
+}
+```
+
+### Verification
+
+```bash
+tsc --noEmit  # Ensure file compiles
+```
+
+---
+
+## Task 2: Consolidate MockStateStore into `test/mocks/mock-state-store.ts`
+
+**Estimated effort**: 1 hour
+**Files created**: `test/mocks/mock-state-store.ts`
+**Files modified**: None (existing vi.mock()-based tests are NOT migrated; this is for new route tests)
+
+### Source
+
+Union of both existing definitions:
+
+- From `test/session-manager.test.ts`: session CRUD methods (`getConfig`, `getSession`, `setSession`, `removeSession`, `getSessions`)
+- From `test/ralph-loop.test.ts`: Ralph state methods (`getConfig`, `getRalphLoopState`, `setRalphLoopState`, `getTasks`, `setTask`, `removeTask`)
+
+### Template
+
+```typescript
+/**
+ * Shared MockStateStore for tests.
+ *
+ * Includes methods for both session management and Ralph loop testing.
+ * All methods are vi.fn() spies — tests can override return values as needed.
+ *
+ * NOTE: This is for direct instantiation in new tests. Existing tests that
+ * use vi.mock('../src/state-store.js') keep their inline definitions.
+ */
+import { vi } from 'vitest';
+
+export class MockStateStore {
+  state: Record<string, unknown> = {
+    sessions: {} as Record<string, unknown>,
+    config: { maxConcurrentSessions: 5 },
+    ralphLoop: { status: 'stopped' },
+    tasks: {} as Record<string, unknown>,
+  };
+
+  // Session methods
+  getConfig = vi.fn(() => this.state.config);
+  getSessions = vi.fn(() => this.state.sessions as Record<string, unknown>);
+  getSession = vi.fn((id: string) => (this.state.sessions as Record<string, unknown>)[id]);
+  setSession = vi.fn((id: string, state: unknown) => {
+    (this.state.sessions as Record<string, unknown>)[id] = state;
+  });
+  removeSession = vi.fn((id: string) => {
+    delete (this.state.sessions as Record<string, unknown>)[id];
+  });
+
+  // Ralph state methods
+  getRalphLoopState = vi.fn(() => this.state.ralphLoop);
+  setRalphLoopState = vi.fn((update: Record<string, unknown>) => {
+    this.state.ralphLoop = { ...(this.state.ralphLoop as Record<string, unknown>), ...update };
+  });
+
+  // Task methods
+  getTasks = vi.fn(() => this.state.tasks);
+  setTask = vi.fn();
+  removeTask = vi.fn();
+
+  // Settings methods
+  getSettings = vi.fn(() => ({}));
+  setSettings = vi.fn();
+
+  // Generic persistence
+  save = vi.fn();
+  load = vi.fn();
+
+  /** Reset all state and mocks for clean test isolation */
+  reset(): void {
+    this.state = {
+      sessions: {},
+      config: { maxConcurrentSessions: 5 },
+      ralphLoop: { status: 'stopped' },
+      tasks: {},
+    };
+    vi.clearAllMocks();
+  }
+}
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+```
+
+---
+
+## Task 3: Create `test/mocks/index.ts` barrel export
+
+**Estimated effort**: 30 minutes
+**Depends on**: Tasks 1, 2
+**Files created**: `test/mocks/index.ts`, `test/mocks/test-helpers.ts`
+**Files modified**: None
+
+### Implementation
+
+1. Create `test/mocks/test-helpers.ts` with the async utilities from `respawn-test-utils.ts`:
+
+```typescript
+/**
+ * Reusable async test helpers.
+ * Extracted from respawn-test-utils.ts.
+ */
+
+/** Wait for an EventEmitter to emit a specific event, with timeout */
+export function waitForEvent(
+  emitter: { once: (event: string, listener: (...args: unknown[]) => void) => void },
+  event: string,
+  timeoutMs = 5000,
+): Promise<unknown> {
+  return new Promise((resolve, reject) => {
+    const timer = setTimeout(
+      () => reject(new Error(`Timed out waiting for event "${event}" after ${timeoutMs}ms`)),
+      timeoutMs,
+    );
+    emitter.once(event, (...args: unknown[]) => {
+      clearTimeout(timer);
+      resolve(args.length === 1 ? args[0] : args);
+    });
+  });
+}
+
+/** Create a deferred promise with external resolve/reject */
+export function createDeferred<T = void>(): {
+  promise: Promise<T>;
+  resolve: (value: T) => void;
+  reject: (reason?: unknown) => void;
+} {
+  let resolve!: (value: T) => void;
+  let reject!: (reason?: unknown) => void;
+  const promise = new Promise<T>((res, rej) => {
+    resolve = res;
+    reject = rej;
+  });
+  return { promise, resolve, reject };
+}
+```
+
+2. Create `test/mocks/index.ts` barrel:
+
+```typescript
+/**
+ * Shared test mocks — import from here instead of defining inline.
+ *
+ * @example
+ * import { MockSession, MockStateStore, terminalOutputs } from './mocks/index.js';
+ */
+
+export { MockSession, createMockSession, terminalOutputs } from './mock-session.js';
+export { MockStateStore } from './mock-state-store.js';
+export { waitForEvent, createDeferred } from './test-helpers.js';
+```
+
+### Verification
+
+```bash
+tsc --noEmit
+```
+
+---
+
+## Task 4: Migrate `respawn-controller.test.ts` to shared mocks
+
+**Estimated effort**: 30 minutes
+**Depends on**: Task 3
+**Files modified**: `test/respawn-controller.test.ts`
+
+### Steps
+
+1. Remove the local `MockSession` class definition (approx. 50 lines).
+2. Add: `import { MockSession } from './mocks/index.js';`
+3. Verify all test methods still exist on the shared mock. The shared mock is a superset, so all existing usage should work.
+4. If the local mock had any test-specific customizations (e.g., extra properties added in `beforeEach`), keep those in the test file as inline assignments on the shared instance.
+5. Run the test to confirm it passes.
+
+### Potential issues
+
+- The local mock's `simulateCompletionMessage()` may have a slightly different output format than the shared mock's (from respawn-test-utils.ts). Verify the respawn controller's completion detection regex matches the shared mock's output pattern (`'\u273b Worked for ...'`).
+- If the local mock adds `pid` or `isWorking` properties that the shared mock doesn't have, add inline assignments in `beforeEach`.
+
+### Verification
+
+```bash
+npx vitest run test/respawn-controller.test.ts
+```
+
+---
+
+## Task 5: Migrate `respawn-team-awareness.test.ts` to shared mocks
+
+**Estimated effort**: 30 minutes
+**Depends on**: Task 3
+**Files modified**: `test/respawn-team-awareness.test.ts`
+
+### Steps
+
+1. Remove the local `MockSession` class definition.
+2. Add: `import { MockSession } from './mocks/index.js';`
+3. Keep `MockTeamWatcher` in this file — it's test-specific and extends the real `TeamWatcher`, not a general-purpose mock.
+4. Run the test to confirm it passes.
+
+### Verification
+
+```bash
+npx vitest run test/respawn-team-awareness.test.ts
+```
+
+---
+
+## Task 6: Create route test scaffold and helpers
+
+**Estimated effort**: 2 hours
+**Depends on**: Task 3
+**Files created**: `test/mocks/mock-route-context.ts`, `test/routes/` directory, `test/routes/_route-test-utils.ts`
+
+### Problem
+
+The 12 route modules in `src/web/routes/` have zero dedicated test coverage. Each route module takes `(app: FastifyInstance, ctx: PortIntersection)` — we need a reusable way to create mock context objects that satisfy the port interfaces.
+
+### Design
+
+Create a `MockRouteContext` factory that builds a mock object satisfying all port interfaces. Each port's methods are `vi.fn()` stubs. Tests can override specific methods as needed.
+
+### Route registration signatures (verified)
+
+Each route module requires a specific port intersection. The mock must satisfy all of them:
+
+| Route Module | Required Ports |
+|-------------|----------------|
+| `registerSessionRoutes` | `SessionPort & EventPort & ConfigPort & InfraPort & AuthPort` |
+| `registerSystemRoutes` | `SessionPort & EventPort & ConfigPort & InfraPort & AuthPort` |
+| `registerRespawnRoutes` | `SessionPort & EventPort & RespawnPort & ConfigPort & InfraPort` |
+| `registerRalphRoutes` | `SessionPort & EventPort & RespawnPort & ConfigPort & InfraPort` |
+| `registerPlanRoutes` | `SessionPort & EventPort & ConfigPort & InfraPort` |
+| `registerCaseRoutes` | `EventPort & ConfigPort` |
+| `registerScheduledRoutes` | `SessionPort & EventPort & InfraPort` |
+| `registerFileRoutes` | `SessionPort` |
+| `registerMuxRoutes` | `InfraPort` |
+| `registerPushRoutes` | `InfraPort` |
+| `registerTeamRoutes` | `InfraPort` |
+| `registerHookEventRoutes` | `EventPort & AuthPort` |
+
+### Implementation
+
+1. Create `test/mocks/mock-route-context.ts`:
+
+```typescript
+/**
+ * Mock context for route handler testing.
+ *
+ * Satisfies ALL port interfaces (SessionPort, EventPort, RespawnPort,
+ * ConfigPort, InfraPort, AuthPort) so any route module can be tested.
+ * Override specific methods in individual tests as needed.
+ *
+ * Verified against actual port interfaces in src/web/ports/:
+ * - SessionPort: 6 methods (sessions, addSession, cleanupSession,
+ *   setupSessionListeners, persistSessionState, persistSessionStateNow,
+ *   getSessionStateWithRespawn)
+ * - EventPort: 5 methods (broadcast, sendPushNotifications, batchTerminalData,
+ *   broadcastSessionStateDebounced, batchTaskUpdate)
+ * - RespawnPort: 2 maps + 4 methods
+ * - ConfigPort: 5 readonly + 7 methods (incl getDefaultClaudeMdPath,
+ *   getLightState, getLightSessionsState, stopTranscriptWatcher)
+ * - InfraPort: 7 readonly + 2 methods (startScheduledRun, stopScheduledRun)
+ * - AuthPort: 3 readonly (authSessions, qrAuthFailures, https)
+ */
+import { vi } from 'vitest';
+import { MockSession, createMockSession } from './mock-session.js';
+
+/**
+ * Creates a mock context that satisfies all port interfaces.
+ * Pre-populated with one session for convenience.
+ */
+export function createMockRouteContext(options?: { sessionId?: string }) {
+  const sessionId = options?.sessionId ?? 'test-session-1';
+  const session = createMockSession(sessionId);
+  const sessions = new Map<string, MockSession>();
+  sessions.set(sessionId, session);
+
+  return {
+    // -- SessionPort --
+    sessions,
+    addSession: vi.fn(),
+    cleanupSession: vi.fn(),
+    setupSessionListeners: vi.fn(),
+    persistSessionState: vi.fn(),
+    persistSessionStateNow: vi.fn(),
+    getSessionStateWithRespawn: vi.fn((s: unknown) => s),
+
+    // -- EventPort --
+    broadcast: vi.fn(),
+    sendPushNotifications: vi.fn(),
+    batchTerminalData: vi.fn(),
+    broadcastSessionStateDebounced: vi.fn(),
+    batchTaskUpdate: vi.fn(),
+
+    // -- RespawnPort --
+    respawnControllers: new Map(),
+    respawnTimers: new Map(),
+    setupRespawnListeners: vi.fn(),
+    setupTimedRespawn: vi.fn(),
+    restoreRespawnController: vi.fn(),
+    saveRespawnConfig: vi.fn(),
+
+    // -- ConfigPort --
+    store: {
+      getConfig: vi.fn(() => ({})),
+      getSessions: vi.fn(() => ({})),
+      getSession: vi.fn(),
+      setSession: vi.fn(),
+      removeSession: vi.fn(),
+      getSettings: vi.fn(() => ({})),
+      setSettings: vi.fn(),
+      getRalphLoopState: vi.fn(() => ({})),
+      setRalphLoopState: vi.fn(),
+      getTasks: vi.fn(() => ({})),
+      save: vi.fn(),
+      load: vi.fn(),
+    },
+    port: 3000,
+    https: false,
+    testMode: true,
+    serverStartTime: Date.now(),
+    getGlobalNiceConfig: vi.fn(async () => undefined),
+    getModelConfig: vi.fn(async () => null),
+    getClaudeModeConfig: vi.fn(async () => ({})),
+    getDefaultClaudeMdPath: vi.fn(async () => undefined),
+    getLightState: vi.fn(() => ({ sessions: [], status: 'ok' })),
+    getLightSessionsState: vi.fn(() => []),
+    startTranscriptWatcher: vi.fn(),
+    stopTranscriptWatcher: vi.fn(),
+
+    // -- InfraPort --
+    mux: {
+      createSession: vi.fn(),
+      killSession: vi.fn(),
+      listSessions: vi.fn(() => []),
+      getStats: vi.fn(() => ({})),
+    },
+    runSummaryTrackers: new Map(),
+    activePlanOrchestrators: new Map(),
+    scheduledRuns: new Map(),
+    teamWatcher: { getTeams: vi.fn(() => []), hasActiveTeammates: vi.fn(() => false) },
+    tunnelManager: null,
+    pushStore: null,
+    startScheduledRun: vi.fn(),
+    stopScheduledRun: vi.fn(),
+
+    // -- AuthPort --
+    authSessions: null,
+    qrAuthFailures: null,
+    // https already declared above in ConfigPort (shared property)
+
+    // Convenience accessors (not part of any port interface)
+    _session: session,
+    _sessionId: sessionId,
+  };
+}
+
+export type MockRouteContext = ReturnType<typeof createMockRouteContext>;
+```
+
+2. Add to `test/mocks/index.ts` barrel:
+
+```typescript
+export { createMockRouteContext, type MockRouteContext } from './mock-route-context.js';
+```
+
+3. Create `test/routes/` directory for route test files.
+
+4. Create `test/routes/_route-test-utils.ts` with Fastify test helpers:
+
+```typescript
+/**
+ * Shared utilities for route testing.
+ *
+ * Creates minimal Fastify instances with just the route module under test
+ * and a mock context. Uses app.inject() for HTTP testing without real ports.
+ */
+import Fastify, { type FastifyInstance } from 'fastify';
+import { createMockRouteContext, type MockRouteContext } from '../mocks/index.js';
+
+export interface RouteTestHarness {
+  app: FastifyInstance;
+  ctx: MockRouteContext;
+}
+
+/**
+ * Creates a Fastify instance with a route module registered against a mock context.
+ *
+ * @param registerFn - The route registration function (e.g., registerSessionRoutes).
+ *   Uses `any` for ctx parameter because route functions expect typed port intersections
+ *   that MockRouteContext satisfies structurally but not nominally.
+ * @param ctxOptions - Optional overrides for the mock context
+ */
+export async function createRouteTestHarness(
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  registerFn: (app: FastifyInstance, ctx: any) => void,
+  ctxOptions?: { sessionId?: string },
+): Promise<RouteTestHarness> {
+  const app = Fastify({ logger: false });
+  const ctx = createMockRouteContext(ctxOptions);
+
+  registerFn(app, ctx);
+  await app.ready();
+
+  return { app, ctx };
+}
+```
+
+### Why `ctx: any` in the harness
+
+Route registration functions like `registerSessionRoutes(app, ctx: SessionPort & EventPort & ConfigPort & InfraPort & AuthPort)` expect specific port intersection types. TypeScript won't accept `unknown` here because it's not assignable to the port types. The `MockRouteContext` satisfies the interfaces structurally (it has all the required properties and methods), but since it's not declared as implementing them, we need `any` at the call site. This is the standard pattern for test mocks in TypeScript.
+
+### Verification
+
+```bash
+tsc --noEmit
+```
+
+---
+
+## Task 7: Add session routes tests
+
+**Estimated effort**: 4 hours
+**Depends on**: Task 6
+**Files created**: `test/routes/session-routes.test.ts`
+**Port**: 3220 (only if SSE tests needed; prefer `app.inject()`)
+
+### Coverage targets
+
+`src/web/routes/session-routes.ts` is the largest route module (43 handlers). Focus on the most critical endpoints first:
+
+#### Priority 1: Session CRUD (must test)
+
+| Method | Path | What to test |
+|--------|------|-------------|
+| `GET` | `/api/sessions` | Returns session list; empty when no sessions |
+| `GET` | `/api/sessions/:id` | Returns session state; 404 for unknown ID |
+| `POST` | `/api/sessions` | Creates session; validates workingDir; rejects invalid paths |
+| `DELETE` | `/api/sessions/:id` | Calls cleanupSession; 404 for unknown ID |
+
+#### Priority 2: Session I/O
+
+| Method | Path | What to test |
+|--------|------|-------------|
+| `POST` | `/api/sessions/:id/input` | Sends input to session; validates input length; 404 for unknown |
+| `POST` | `/api/sessions/:id/resize` | Validates cols/rows bounds; 404 for unknown |
+| `GET` | `/api/sessions/:id/buffer` | Returns terminal buffer; 404 for unknown |
+
+#### Priority 3: Session actions
+
+| Method | Path | What to test |
+|--------|------|-------------|
+| `POST` | `/api/sessions/:id/run` | Runs prompt on session |
+| `POST` | `/api/sessions/:id/clear` | Clears session |
+| `POST` | `/api/sessions/:id/compact` | Compacts session |
+| `POST` | `/api/sessions/:id/interactive` | Starts interactive mode |
+| `POST` | `/api/sessions/:id/quick-start` | Quick start flow |
+
+### Test pattern
+
+```typescript
+import { describe, it, expect, beforeEach, afterEach } from 'vitest';
+import { createRouteTestHarness, type RouteTestHarness } from './_route-test-utils.js';
+import { registerSessionRoutes } from '../../src/web/routes/session-routes.js';
+
+describe('session-routes', () => {
+  let harness: RouteTestHarness;
+
+  beforeEach(async () => {
+    harness = await createRouteTestHarness(registerSessionRoutes);
+  });
+
+  afterEach(async () => {
+    await harness.app.close();
+  });
+
+  describe('GET /api/sessions', () => {
+    it('returns empty array when no sessions', async () => {
+      harness.ctx.sessions.clear();
+      const res = await harness.app.inject({ method: 'GET', url: '/api/sessions' });
+      expect(res.statusCode).toBe(200);
+      expect(JSON.parse(res.body)).toEqual([]);
+    });
+
+    it('returns session list with one session', async () => {
+      const res = await harness.app.inject({ method: 'GET', url: '/api/sessions' });
+      expect(res.statusCode).toBe(200);
+      const sessions = JSON.parse(res.body);
+      expect(sessions).toHaveLength(1);
+    });
+  });
+
+  describe('GET /api/sessions/:id', () => {
+    it('returns 404 for unknown session', async () => {
+      const res = await harness.app.inject({
+        method: 'GET',
+        url: '/api/sessions/nonexistent',
+      });
+      expect(res.statusCode).toBe(404);
+    });
+  });
+
+  describe('POST /api/sessions/:id/input', () => {
+    it('rejects input exceeding max length', async () => {
+      const res = await harness.app.inject({
+        method: 'POST',
+        url: `/api/sessions/${harness.ctx._sessionId}/input`,
+        payload: { input: 'x'.repeat(65537) },
+      });
+      expect(res.statusCode).toBe(400);
+    });
+  });
+
+  describe('POST /api/sessions/:id/resize', () => {
+    it('rejects cols exceeding max', async () => {
+      const res = await harness.app.inject({
+        method: 'POST',
+        url: `/api/sessions/${harness.ctx._sessionId}/resize`,
+        payload: { cols: 501, rows: 24 },
+      });
+      expect(res.statusCode).toBe(400);
+    });
+  });
+});
+```
+
+### Key assertions to include
+
+- **404 for unknown sessions**: Every `:id` endpoint must return 404 for nonexistent IDs
+- **Input validation**: Bad paths, oversized inputs, invalid resize dimensions
+- **Side effects**: Verify `ctx.broadcast()` was called with correct event type after mutations
+- **Response shape**: Verify response bodies match expected API types
+
+### Verification
+
+```bash
+npx vitest run test/routes/session-routes.test.ts
+```
+
+---
+
+## Task 8: Add system + respawn routes tests
+
+**Estimated effort**: 4 hours
+**Depends on**: Task 6
+**Files created**: `test/routes/system-routes.test.ts`, `test/routes/respawn-routes.test.ts`
+
+### System routes (`src/web/routes/system-routes.ts`)
+
+Focus on status and configuration endpoints:
+
+| Method | Path | What to test |
+|--------|------|-------------|
+| `GET` | `/api/status` | Returns server status with uptime, session count |
+| `GET` | `/api/stats` | Returns mux stats |
+| `GET` | `/api/config` | Returns current config |
+| `PUT` | `/api/config` | Updates config; validates input |
+| `GET` | `/api/settings` | Returns user settings |
+| `PUT` | `/api/settings` | Updates settings; validates input |
+| `GET` | `/api/subagents` | Returns subagent list |
+| `GET` | `/api/screenshots` | Returns screenshot list |
+
+### Respawn routes (`src/web/routes/respawn-routes.ts`)
+
+| Method | Path | What to test |
+|--------|------|-------------|
+| `GET` | `/api/sessions/:id/respawn` | Returns respawn status; null when not configured |
+| `POST` | `/api/sessions/:id/respawn/start` | Starts respawn; 404 for unknown session |
+| `POST` | `/api/sessions/:id/respawn/stop` | Stops respawn; 404 for unknown session |
+| `PUT` | `/api/sessions/:id/respawn/config` | Updates respawn config; validates |
+| `POST` | `/api/sessions/:id/respawn/enable` | Enables respawn loop |
+| `POST` | `/api/sessions/:id/respawn/disable` | Disables respawn loop |
+
+### Test patterns
+
+Same pattern as Task 7 — `createRouteTestHarness` with `registerSystemRoutes` / `registerRespawnRoutes`.
+
+For respawn tests, pre-populate `ctx.respawnControllers` with a mock controller in `beforeEach`:
+
+```typescript
+beforeEach(async () => {
+  harness = await createRouteTestHarness(registerRespawnRoutes);
+  // Add a mock respawn controller for the default session
+  harness.ctx.respawnControllers.set(harness.ctx._sessionId, {
+    getState: vi.fn(() => 'idle'),
+    getConfig: vi.fn(() => ({})),
+    getStatus: vi.fn(() => ({ state: 'idle', health: 100 })),
+    start: vi.fn(),
+    stop: vi.fn(),
+    updateConfig: vi.fn(),
+    enable: vi.fn(),
+    disable: vi.fn(),
+  });
+});
+```
+
+### Verification
+
+```bash
+npx vitest run test/routes/system-routes.test.ts
+npx vitest run test/routes/respawn-routes.test.ts
+```
+
+---
+
+## Task 9: Slim down `respawn-test-utils.ts`
+
+**Estimated effort**: 30 minutes
+**Depends on**: Tasks 4, 5
+**Files modified**: `test/respawn-test-utils.ts`
+
+After Tasks 4–5 are verified passing with shared mocks, slim down `respawn-test-utils.ts` to remove duplicates.
+
+### Steps
+
+1. **Remove** from `respawn-test-utils.ts` what has been moved to shared mocks:
+   - `MockSession` class → now in `test/mocks/mock-session.ts`
+   - `createMockSession()` → now in `test/mocks/mock-session.ts`
+   - `terminalOutputs` → now in `test/mocks/mock-session.ts`
+   - `waitForEvent()` / `createDeferred()` → now in `test/mocks/test-helpers.ts`
+
+2. **Keep** respawn-specific utilities that don't belong in the general mocks:
+   - `TimeController` / `createTimeController()` — respawn-specific timer control
+   - `MockAiIdleChecker` / `MockAiPlanChecker` — respawn-specific AI mocks
+   - `createStateTracker()` / `createEventRecorder()` — respawn state tracking
+   - `FAST_TEST_CONFIG` / `AI_ENABLED_TEST_CONFIG` — respawn config presets
+   - `waitForState()` — respawn state machine waiter
+
+3. **Update imports** in `respawn-test-utils.ts` to re-use shared mocks:
+   ```typescript
+   import { MockSession, createMockSession, terminalOutputs } from './mocks/index.js';
+   import { waitForEvent, createDeferred } from './mocks/index.js';
+   export { MockSession, createMockSession, terminalOutputs, waitForEvent, createDeferred };
+   ```
+
+This preserves backward compatibility for any future tests that import from `respawn-test-utils.ts` directly while eliminating the duplication.
+
+### Verification
+
+```bash
+tsc --noEmit
+npx vitest run test/respawn-controller.test.ts
+npx vitest run test/respawn-team-awareness.test.ts
+```
+
+---
+
+## What is NOT in scope (and why)
+
+### Migrating `session-manager.test.ts` and `ralph-loop.test.ts` mocks
+
+Both files define mocks inside `vi.mock()` factories that replace entire modules:
+
+```typescript
+// session-manager.test.ts — mock replaces ../src/session.js
+vi.mock('../src/session.js', () => {
+  class MockSession extends EventEmitter { ... }
+  return { Session: MockSession };
+});
+
+// ralph-loop.test.ts — mock replaces ../src/state-store.js
+vi.mock('../src/state-store.js', () => {
+  class MockStateStore { ... }
+  return { getStore: vi.fn(() => instance), StateStore: MockStateStore };
+});
+```
+
+These are fundamentally different from the direct-instantiation pattern:
+- The `vi.mock()` factory runs in an isolated scope — outer imports are not available
+- The mock class must be returned with the exact export names (`Session`, `getStore`, `StateStore`)
+- The `session-manager.test.ts` MockSession auto-registers into a shared `mockState.sessions` Map (tight coupling with test setup)
+
+Migrating would require `vi.hoisted()` to share the class between factory and test scope, plus restructuring the test's module-mocking setup. This is high-complexity, high-risk refactoring with limited benefit since these tests already work. The shared `MockStateStore` in `test/mocks/` is available for **new** tests (like route tests) that use direct instantiation instead.
+
+### Full integration tests with real Fastify server
+
+Route tests use `app.inject()` which simulates HTTP without opening ports. Full integration tests that spin up `WebServer`, create real sessions, and stream SSE would be valuable but are a separate effort requiring:
+- A test WebServer factory
+- Session lifecycle management in tests
+- SSE client test utilities
+- Significantly more setup/teardown complexity
+
+### Testing auth middleware in route tests
+
+Route tests bypass authentication (no auth middleware registered on the test Fastify instance). Auth middleware has its own dedicated tests in `auth-security.test.ts` and `qr-auth.test.ts`. Testing auth + routes together is a future integration test concern.
+
+### Testing SSE event streaming
+
+SSE integration requires a running server with `EventSource` client. This is significantly more complex than `app.inject()` tests and is deferred. The existing `sse-events.test.ts` covers SSE patterns.
+
+### Complete route coverage for all 12 modules
+
+This phase covers the 3 highest-value route modules (session, system, respawn — 98 of 162 handlers). The remaining 9 modules (ralph, plan, push, team, mux, file, scheduled, hook-event, case) should be added incrementally in follow-up work.
+
+---
+
+## Summary
+
+| Metric | Before | After |
+|--------|--------|-------|
+| MockSession definitions | 4 (across 4 files) | 1 shared (2 vi.mock() copies remain, intentionally) |
+| MockStateStore definitions | 2 (across 2 files) | 1 shared (2 vi.mock() copies remain, intentionally) |
+| Files importing from `respawn-test-utils.ts` | 0 | Utilities split into `test/mocks/` |
+| Route test files | 0 | 3 (session, system, respawn) |
+| Route handlers with dedicated tests | 0 | ~30 (highest-priority endpoints) |
+| Shared mock directory | None | `test/mocks/` with 5 files + barrel |
+
+### Final verification checklist
+
+```bash
+# Type checking
+tsc --noEmit
+
+# Linting
+npm run lint
+
+# Formatting
+npm run format:check
+
+# Run all affected tests individually
+npx vitest run test/respawn-controller.test.ts
+npx vitest run test/respawn-team-awareness.test.ts
+npx vitest run test/routes/session-routes.test.ts
+npx vitest run test/routes/system-routes.test.ts
+npx vitest run test/routes/respawn-routes.test.ts
+
+# Verify unchanged tests still pass
+npx vitest run test/session-manager.test.ts
+npx vitest run test/ralph-loop.test.ts
+
+# Dev server still starts
+npx tsx src/index.ts web --port 3099 &
+curl -s http://localhost:3099/api/status | jq .status  # "ok"
+kill %1
+```

docs/qr-auth-plan.md

@@ -0,0 +1,723 @@
+# QR Code Authentication Plan
+
+> Ephemeral, single-use auth tokens embedded in the tunnel QR code — scan to auto-authenticate, while the bare tunnel URL stays password-protected.
+
+## Problem
+
+When the Cloudflare tunnel is active, anyone who discovers the `*.trycloudflare.com` URL can access Codeman (they just need the Basic Auth password, or if no password is set, full open access). The QR code currently encodes the raw tunnel URL — it provides no additional security. We want:
+
+1. **Scanning the QR code** → seamless, instant access (no password prompt)
+2. **Having only the URL** → blocked by Basic Auth (no access without credentials)
+
+## Design
+
+### Core Concept: Ephemeral Single-Use QR Tokens
+
+The server maintains a rotating pool of short-lived, single-use tokens. The QR code encodes a short URL containing a lookup code that maps to the real token server-side. When scanned, the server validates the token, atomically consumes it, issues a session cookie, and redirects to `/`. The token is **not** the password — it's a separate, independent, ephemeral authentication pathway.
+
+```
+Desktop   → displays QR (auto-refreshes every 60s via SSE)
+QR Code   → https://abc-xyz.trycloudflare.com/q/Xk9mQ3
+Phone     → scans, GET /q/Xk9mQ3
+Server    → looks up short code via Map (hash-based, timing-safe)
+          → finds token record → validates TTL
+          → atomically consumes token (single-use)
+          → issues codeman_session cookie
+          → 302 redirect to /
+          → SSE push: new QR with embedded SVG for desktop display
+          → desktop toast: "Device [IP] authenticated via QR"
+          → audit log entry to session-lifecycle.jsonl
+User      → lands on app, fully authenticated
+```
+
+Someone who only has `https://abc-xyz.trycloudflare.com/` gets the standard Basic Auth prompt.
+
+### Token Properties
+
+| Property | Value |
+|----------|-------|
+| Length | 32 bytes (256 bits entropy) |
+| Generation | `crypto.randomBytes(32).toString('hex')` |
+| Short code | 6 chars base62, rejection-sampled (no modulo bias) |
+| Short code derivation | Independent random generation (not derived from token) |
+| Storage | In-memory `Map<shortCode, QrTokenRecord>` (no disk persistence) |
+| TTL | 60 seconds (auto-rotation via timer), 90s grace for previous token |
+| Effective window | Up to 90 seconds for the previous token (documented, not hidden) |
+| Usage | **Single-use** — atomically consumed on first valid scan |
+| URL format | Short code in path (`/q/Xk9mQ3`), not query params |
+| URL length | ~53-56 chars total — targets QR Version 4 (33x33) for fast scanning |
+| Scope | Only valid when `CODEMAN_PASSWORD` is set (no point without auth) |
+| Lookup | `Map.get()` — hash-based O(1), no timing side-channel |
+
+### Why This Design?
+
+**Why not embed the password directly?**
+- Password would appear in browser history, Cloudflare edge logs, and URL bars
+- Password can't be rotated independently from QR access
+
+**Why not a long-lived multi-use token? (original design)**
+- A static token is functionally a second password — if the QR image leaks (screenshot shared, shoulder surfing, Cloudflare logs), the attacker has permanent access
+- The USENIX Security 2025 paper ["Demystifying the (In)Security of QR Code-based Login"](https://www.usenix.org/conference/usenixsecurity25/presentation/zhang-xin) found 47 of the top-100 websites vulnerable due to exactly this pattern — missing single-use enforcement and long-lived tokens were 2 of the 6 critical design flaws identified
+
+**Why short codes in the URL path instead of query params?**
+- Query params (`?t=TOKEN`) leak into browser history, address bar, `Referer` headers, and Cloudflare edge logs
+- Path-based short codes (`/q/Xk9mQ3`) are opaque references — the real token never appears in URLs
+- Short codes are 6-char base62 (62^6 = 56.8 billion combinations), sufficient for lookup since they're backed by the full 256-bit token for validation and rate-limited to 10 attempts/IP
+- The short `/q/` path (vs `/qr-auth/`) saves 7 bytes, helping keep the QR at Version 4 (33x33 modules) instead of Version 5 (37x37) — faster scanning on budget phones
+
+## Auth Flow Diagram
+
+```
+┌─────────────┐    scan QR     ┌──────────────────────────────────────┐
+│  Mobile      │ ────────────→ │ GET /q/Xk9mQ3                        │
+│  Device      │               │                                      │
+└─────────────┘               │ 1. Auth middleware sees /q/            │
+                               │    → skips Basic Auth check           │
+                               │ 2. Route handler: Map.get(shortCode)  │
+                               │    → hash-based lookup (timing-safe)  │
+                               │ 3. Checks TTL (90s grace for prev)    │
+                               │    → token not expired?               │
+                               │ 4. Checks consumed flag               │
+                               │    → not already used?                 │
+                               │ 5. Atomically marks token consumed    │
+                               │ 6. Issues codeman_session cookie      │
+                               │ 7. 302 redirect to /                  │
+                               │ 8. Audit log → session-lifecycle.jsonl│
+                               │ 9. SSE push: tunnel:qrRegenerated     │
+                               │    → desktop refreshes QR (SVG inline)│
+                               │ 10. Desktop toast: "Device auth'd"    │
+                               └──────────────────────────────────────┘
+
+┌─────────────┐  replay URL   ┌──────────────────────────────────────┐
+│  Attacker    │ ────────────→ │ GET /q/Xk9mQ3                        │
+│ (stale code) │               │                                      │
+└─────────────┘               │ 1. Map.get(shortCode) → not found     │
+                               │    OR token consumed OR expired       │
+                               │ 2. Increment QR rate limit counter    │
+                               │    (separate from Basic Auth counter) │
+                               │ 3. 401 Unauthorized                   │
+                               └──────────────────────────────────────┘
+
+┌─────────────┐   URL only    ┌──────────────────────────────────────┐
+│  Attacker    │ ────────────→ │ GET /                                │
+│  (no token)  │               │                                      │
+└─────────────┘               │ 1. Auth middleware checks cookie      │
+                               │    → no cookie                        │
+                               │ 2. Checks Basic Auth header           │
+                               │    → no header                        │
+                               │ 3. Returns 401 + WWW-Authenticate    │
+                               │    → Browser shows password popup     │
+                               └──────────────────────────────────────┘
+```
+
+## Implementation
+
+### 1. Token Manager — `src/tunnel-manager.ts`
+
+Add a `QrTokenRecord` type and token rotation logic to `TunnelManager`. The token rotates every 60 seconds. A consumed token is immediately replaced. Up to 2 tokens can be valid simultaneously (current + previous, to handle the race where someone scans right as rotation happens). The previous token has a 90s grace period (not a full extra 60s — only enough to cover the scan-during-rotation race).
+
+**Design decisions from security review:**
+- **Map-based lookup** (not array scan) — `Map.get()` uses hash-based O(1) lookup, eliminating timing side-channels from string comparison
+- **Rejection sampling** for short codes — avoids modulo bias (`256 % 62 != 0` gives 25% overrepresentation for first 6 charset chars)
+- **SVG cache** — stores generated QR SVG per rotation cycle to avoid regenerating on every `/api/tunnel/qr` poll
+- **Separate rate limit counter** — QR auth failures tracked independently from Basic Auth failures
+
+```typescript
+import { randomBytes } from 'node:crypto';
+
+interface QrTokenRecord {
+  token: string;       // 64 hex chars (256 bits)
+  shortCode: string;   // 6 chars base62 (for URL path)
+  createdAt: number;   // Date.now()
+  consumed: boolean;   // single-use flag
+}
+
+const QR_TOKEN_TTL_MS = 60_000;        // 60 seconds
+const QR_TOKEN_GRACE_MS = 90_000;      // 90s grace for previous token (scan-during-rotation)
+const SHORT_CODE_LENGTH = 6;
+const QR_RATE_LIMIT_MAX = 30;          // global rate limit across all IPs
+const QR_RATE_LIMIT_WINDOW_MS = 60_000; // 1 minute window
+
+/** Rejection-sampled short code generation — no modulo bias */
+function generateShortCode(): string {
+  const chars = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789';
+  const maxUnbiased = 248; // largest multiple of 62 that fits in a byte (248 = 62 * 4)
+  const result: string[] = [];
+  while (result.length < SHORT_CODE_LENGTH) {
+    const [byte] = randomBytes(1);
+    if (byte < maxUnbiased) result.push(chars[byte % 62]);
+    // else: discard and re-draw (rejection sampling)
+  }
+  return result.join('');
+}
+
+export class TunnelManager extends EventEmitter {
+  // Map-based lookup: shortCode → QrTokenRecord (timing-safe, no string comparison)
+  private qrTokensByCode = new Map<string, QrTokenRecord>();
+  private currentShortCode: string | null = null;
+  private rotationTimer: ReturnType<typeof setInterval> | null = null;
+
+  // SVG cache — regenerated only on token rotation, not per request
+  private cachedQrSvg: { shortCode: string; svg: string } | null = null;
+
+  // Global rate limit counter (separate from Basic Auth rate limiting)
+  private qrAttemptCount = 0;
+  private qrRateLimitResetTimer: ReturnType<typeof setInterval> | null = null;
+
+  constructor() {
+    super();
+    this.rotateToken();
+    this.rotationTimer = setInterval(() => this.rotateToken(), QR_TOKEN_TTL_MS);
+    this.qrRateLimitResetTimer = setInterval(() => { this.qrAttemptCount = 0; }, QR_RATE_LIMIT_WINDOW_MS);
+  }
+
+  private rotateToken(): void {
+    const record: QrTokenRecord = {
+      token: randomBytes(32).toString('hex'),
+      shortCode: generateShortCode(),
+      createdAt: Date.now(),
+      consumed: false,
+    };
+
+    // Evict expired tokens from the Map
+    const now = Date.now();
+    for (const [code, rec] of this.qrTokensByCode) {
+      if (now - rec.createdAt > QR_TOKEN_GRACE_MS || rec.consumed) {
+        this.qrTokensByCode.delete(code);
+      }
+    }
+
+    this.qrTokensByCode.set(record.shortCode, record);
+    this.currentShortCode = record.shortCode;
+    this.cachedQrSvg = null; // invalidate SVG cache
+    this.emit('qrTokenRotated');
+  }
+
+  /** Get the current (newest) token's short code for QR URL */
+  getCurrentShortCode(): string | undefined {
+    return this.currentShortCode ?? undefined;
+  }
+
+  /** Get cached QR SVG, regenerating only if the short code changed */
+  async getQrSvg(tunnelUrl: string): Promise<string> {
+    const code = this.currentShortCode;
+    if (!code) throw new Error('No QR token available');
+    if (this.cachedQrSvg?.shortCode === code) return this.cachedQrSvg.svg;
+    const QRCode = require('qrcode');
+    const svg = await QRCode.toString(`${tunnelUrl}/q/${code}`, { type: 'svg', margin: 2, width: 256 });
+    this.cachedQrSvg = { shortCode: code, svg };
+    return svg;
+  }
+
+  /**
+   * Validate and atomically consume a token by short code.
+   * Returns { success, ip?, ua? } for audit logging on success.
+   * Map.get() is hash-based — no timing side-channel from string comparison.
+   */
+  consumeToken(shortCode: string): boolean {
+    // Global rate limit (across all IPs)
+    if (this.qrAttemptCount >= QR_RATE_LIMIT_MAX) return false;
+    this.qrAttemptCount++;
+
+    const record = this.qrTokensByCode.get(shortCode);
+    if (!record) return false;
+    if (record.consumed) return false;
+
+    const now = Date.now();
+    if (now - record.createdAt > QR_TOKEN_GRACE_MS) return false;
+
+    // Atomic consume (single-threaded JS = no race)
+    record.consumed = true;
+    // Immediately rotate so desktop gets a fresh QR
+    this.rotateToken();
+    this.emit('qrTokenRegenerated');
+    return true;
+  }
+
+  /** Force-regenerate (manual revocation via API) */
+  regenerateQrToken(): void {
+    // Invalidate all existing tokens
+    this.qrTokensByCode.clear();
+    this.currentShortCode = null;
+    this.rotateToken();
+    this.emit('qrTokenRegenerated');
+  }
+
+  stopRotation(): void {
+    if (this.rotationTimer) {
+      clearInterval(this.rotationTimer);
+      this.rotationTimer = null;
+    }
+    if (this.qrRateLimitResetTimer) {
+      clearInterval(this.qrRateLimitResetTimer);
+      this.qrRateLimitResetTimer = null;
+    }
+  }
+}
+```
+
+### 2. Auth Middleware Bypass — `src/web/middleware/auth.ts`
+
+Add `/q/` to the bypass list (same pattern as `/api/hook-event`). The route handler itself handles token validation and rate limiting.
+
+```typescript
+// In the onRequest hook, add before Basic Auth check:
+if (req.url.startsWith('/q/')) {
+  done();  // Let the route handler deal with token validation
+  return;
+}
+```
+
+**Important**: Unlike `/api/hook-event` (localhost-only), `/q/` must be reachable from any IP (remote devices scan the QR). Rate limiting is handled by two independent mechanisms:
+1. **Per-IP rate limit** — reuses the `authFailures` StaleExpirationMap (10 attempts/IP/15min), but tracked via a **separate counter** from Basic Auth failures (so a user who fat-fingers their password doesn't burn their QR attempts)
+2. **Global path rate limit** — `TunnelManager.qrAttemptCount` caps total QR attempts to 30/minute across all IPs, defending against distributed brute force
+
+### 3. Auto-Auth Route — `src/web/routes/system-routes.ts`
+
+Add `GET /q/:code` as a top-level route (not under `/api/`):
+
+```typescript
+app.get('/q/:code', async (req, reply) => {
+  const shortCode = (req.params as { code: string }).code;
+  const authPassword = process.env.CODEMAN_PASSWORD;
+
+  // No point if auth isn't enabled
+  if (!authPassword) {
+    return reply.redirect('/');
+  }
+
+  // Per-IP rate limit (separate counter from Basic Auth failures)
+  const clientIp = req.ip;
+  const qrFailures = ctx.authState.qrAuthFailures?.get(clientIp) ?? 0;
+  if (qrFailures >= 10) {
+    return reply.code(429).send('Too Many Requests');
+  }
+
+  // Validate and atomically consume the token
+  // consumeToken() also checks the global rate limit (30/min across all IPs)
+  if (!shortCode || !ctx.tunnelManager.consumeToken(shortCode)) {
+    ctx.authState.qrAuthFailures?.set(clientIp, qrFailures + 1);
+    return reply.code(401).send('Invalid or expired QR code');
+  }
+
+  // Issue session cookie (same as Basic Auth success path)
+  const sessionToken = randomBytes(32).toString('hex');
+  const clientUA = req.headers['user-agent'] ?? '';
+  ctx.authState.authSessions?.set(sessionToken, {
+    ip: clientIp,
+    ua: clientUA,
+    createdAt: Date.now(),
+  });
+  ctx.authState.qrAuthFailures?.delete(clientIp);
+
+  // Audit log — write to session-lifecycle.jsonl for forensic analysis
+  ctx.lifecycleLog?.append({
+    event: 'qr_auth',
+    ip: clientIp,
+    ua: clientUA,
+    timestamp: Date.now(),
+    shortCodePrefix: shortCode.slice(0, 3) + '***', // partial for privacy
+  });
+
+  reply.setCookie(AUTH_COOKIE_NAME, sessionToken, {
+    httpOnly: true,
+    secure: ctx.https,
+    sameSite: 'lax',
+    maxAge: 86400,  // 24h
+    path: '/',
+  });
+
+  // Broadcast auth notification — desktop sees who authenticated (QRLjacking detection)
+  broadcast('tunnel:qrAuthUsed', {
+    ip: clientIp,
+    ua: clientUA,
+    timestamp: Date.now(),
+  });
+
+  return reply.redirect('/');
+});
+```
+
+### 4. Update QR Code URL — `src/web/routes/system-routes.ts`
+
+Modify `/api/tunnel/qr` to encode the short-code URL. Uses the `TunnelManager.getQrSvg()` cache — SVG is regenerated only when the token rotates, not on every request.
+
+```typescript
+app.get('/api/tunnel/qr', async (_req, reply) => {
+  const url = ctx.tunnelManager.getUrl();
+  if (!url) {
+    return reply.code(404).send(createErrorResponse(ApiErrorCode.NOT_FOUND, 'Tunnel not running'));
+  }
+
+  const authPassword = process.env.CODEMAN_PASSWORD;
+
+  // If auth is enabled, use the cached SVG with embedded short code
+  if (authPassword) {
+    const svg = await ctx.tunnelManager.getQrSvg(url);
+    return { svg, authEnabled: true };
+  }
+
+  // No auth — just encode the raw tunnel URL
+  const QRCode = require('qrcode');
+  const svg = await QRCode.toString(url, { type: 'svg', margin: 2, width: 256 });
+  return { svg, authEnabled: false };
+});
+```
+
+### 5. Token Regeneration Endpoint — `src/web/routes/system-routes.ts`
+
+Manual revocation — invalidates ALL existing tokens and creates a fresh one:
+
+```typescript
+app.post('/api/tunnel/qr/regenerate', async () => {
+  ctx.tunnelManager.regenerateQrToken();
+  return { success: true };
+});
+```
+
+### 6. Frontend Updates — `src/web/public/app.js`
+
+#### QR Overlay Changes
+
+- **Auto-refresh via inline SVG**: Listen for `tunnel:qrRotated` SSE events which now include the SVG directly in the payload — no extra HTTP fetch needed, sub-50ms refresh on desktop.
+- **Countdown indicator**: Small "expires in Xs" text under the QR that counts down from 60. Reassures the user the QR is live and not stale.
+- **Regenerate button**: "Regenerate QR" button. Calls `POST /api/tunnel/qr/regenerate` — SSE event delivers the new SVG.
+- **Auth badge**: Lock icon or "Single-use auth" label when auth is active.
+- **URL display**: Show the raw tunnel URL (not the auth URL) for manual copy — users who copy the URL authenticate via Basic Auth. The QR is the fast path.
+- **Auth notification toast**: When `tunnel:qrAuthUsed` fires, show a 10-second toast: "Device [IP] authenticated via QR (Safari). Not you? [Revoke]". This is the primary QRLjacking detection mechanism (USENIX Flaw-5).
+
+```javascript
+// Auto-refresh QR on rotation — SVG is inline in the event payload
+addListener('tunnel:qrRotated', (data) => {
+  if (data.svg) {
+    updateQrDisplay(data.svg);  // direct DOM update, no fetch
+  } else {
+    refreshTunnelQR();          // fallback: fetch from API
+  }
+});
+
+// Also refresh on manual regeneration
+addListener('tunnel:qrRegenerated', (data) => {
+  if (data.svg) {
+    updateQrDisplay(data.svg);
+  } else {
+    refreshTunnelQR();
+  }
+});
+
+// QRLjacking detection — notify desktop user when QR is consumed
+addListener('tunnel:qrAuthUsed', (data) => {
+  showNotificationToast(
+    `Device authenticated via QR (${parseUAFamily(data.ua)}, ${data.ip}). Not you?`,
+    {
+      duration: 10000,
+      action: { label: 'Revoke', onClick: () => revokeAllSessions() },
+    }
+  );
+});
+
+// In showTunnelQR(), after fetching /api/tunnel/qr:
+if (data.authEnabled) {
+  const badge = document.createElement('div');
+  badge.textContent = 'Single-use auth \u00b7 refreshes every 60s';
+  badge.style.cssText = 'margin-top:8px;font-size:11px;color:var(--text-secondary)';
+  container.parentElement.appendChild(badge);
+}
+```
+
+#### Welcome Screen QR
+
+Same auto-refresh behavior applies to `_updateWelcomeTunnelBtn()` — the QR is fetched from `/api/tunnel/qr` so token embedding happens automatically.
+
+### 7. SSE Events
+
+Three events for the frontend. QR rotation events embed the SVG directly in the payload to eliminate an extra HTTP fetch — the desktop gets the new QR in a single SSE push (~2-5KB SVG, well within SSE limits).
+
+```typescript
+// In server.ts, listen for tunnelManager events:
+
+// Auto-rotation every 60s — desktop refreshes QR silently (SVG inline)
+tunnelManager.on('qrTokenRotated', async () => {
+  const url = tunnelManager.getUrl();
+  if (url && process.env.CODEMAN_PASSWORD) {
+    const svg = await tunnelManager.getQrSvg(url);
+    broadcast('tunnel:qrRotated', { svg });
+  } else {
+    broadcast('tunnel:qrRotated', {});
+  }
+});
+
+// Manual regeneration or post-consumption — desktop refreshes QR (SVG inline)
+tunnelManager.on('qrTokenRegenerated', async () => {
+  const url = tunnelManager.getUrl();
+  if (url && process.env.CODEMAN_PASSWORD) {
+    const svg = await tunnelManager.getQrSvg(url);
+    broadcast('tunnel:qrRegenerated', { svg });
+  } else {
+    broadcast('tunnel:qrRegenerated', {});
+  }
+});
+
+// QR auth consumed — desktop shows notification toast (QRLjacking detection)
+// Note: this is broadcast from the route handler, not tunnelManager
+// Event: tunnel:qrAuthUsed { ip, ua, timestamp }
+```
+
+### 8. Session Cookie Binding & Revocation
+
+Enhance session records to include device context for audit purposes. The UA is stored for **logging only** — not for blocking.
+
+**Why no UA-family blocking (`majorUAChanged`)?** Security review found this is security theater:
+- UA strings are trivially spoofable by any attacker who can steal a cookie
+- Chrome UA reduction (2022+) makes family detection unreliable
+- Mobile WebView → browser switches trigger false positives on the same device
+- HttpOnly + Secure + SameSite=lax + 24h TTL already protect against cookie theft
+- The attacker who can exfiltrate a cookie can also replay the exact UA
+
+Instead, provide **manual session revocation** as the active defense:
+
+```typescript
+// Session record stores device context for audit logging (not blocking):
+ctx.authState.authSessions?.set(sessionToken, {
+  ip: clientIp,
+  ua: req.headers['user-agent'] ?? '',
+  createdAt: Date.now(),
+  method: 'qr',  // 'qr' | 'basic' — tracks how session was created
+});
+
+// Manual revocation endpoint — kill specific session or all sessions
+app.post('/api/auth/revoke', async (req, reply) => {
+  const { sessionToken: target } = req.body as { sessionToken?: string };
+  if (target) {
+    ctx.authState.authSessions?.delete(target);
+  } else {
+    // Revoke all sessions (nuclear option)
+    ctx.authState.authSessions?.clear();
+  }
+  return { success: true };
+});
+```
+
+**Note**: This is a breaking type change. The `AuthState` interface must be updated from `StaleExpirationMap<string, string>` (token → clientIp) to `StaleExpirationMap<string, { ip, ua, createdAt, method }>`. All session validation code in `auth.ts` must be updated simultaneously.
+
+### 9. Cleanup — `src/tunnel-manager.ts`
+
+Stop the rotation timer in the `stop()` method:
+
+```typescript
+stop(): void {
+  this.stopRotation();
+  // ... existing cleanup
+}
+```
+
+## Security Analysis
+
+### Threat Model
+
+| Threat | Attack Vector | Mitigation | Residual Risk |
+|--------|--------------|------------|---------------|
+| **QR screenshot shared** | Attacker gets image of QR code | Single-use: token consumed on first scan. 60s TTL: expired by the time attacker tries. Desktop toast notification alerts user if someone else scans. | If attacker scans faster than legitimate user (~seconds), they win the race. Low risk: requires physical proximity + speed. User sees notification and can revoke. |
+| **Cloudflare edge logs** | Cloudflare logs the full URL path | Short code is opaque (6-char lookup key), not the real token. Single-use: replaying from logs always fails. 60s TTL (90s grace): expired before log review. `trycloudflare.com` quick tunnels have no customer-accessible logging controls — the privacy implications are inherent to using free quick tunnels. | Cloudflare has TLS termination access regardless. Ephemeral short codes are far less valuable than a permanent token. |
+| **Brute force short code** | Attacker guesses `/q/XXXXXX` | Per-IP rate limiting (10/IP/15min) + global path rate limit (30/min across all IPs). 62^6 = 56.8 billion combinations. Only ~2 valid codes at any time. | Infeasible: expected guesses to hit = ~2.8×10^10, rate limits block well before. |
+| **Replay attack** | Reuse a previously valid URL | Single-use consumption + 60s TTL (90s grace). Old codes always 401. | None — replay is impossible by design. |
+| **QRLjacking** | Attacker displays your QR on phishing site | No companion app = limited mitigation. However: 60s rotation means attacker must relay in real-time. Desktop toast notification ("Device [IP] authenticated via QR. Not you? [Revoke]") provides real-time detection. Self-hosted single-user context makes phishing implausible. | Theoretical risk for multi-user deployments. Mitigated by notification toast for single-user. Note: Signal's linked-device QR flow was exploited by Russian state actors (UNC5792/Sandworm) via quishing in 2025 — but that targeted a multi-user messaging platform, not a self-hosted dev tool. |
+| **Session cookie theft** | XSS or network sniffing steals cookie | HttpOnly + Secure flags. SameSite=lax prevents CSRF. 24h TTL limits exposure window. Manual revocation via `/api/auth/revoke`. | Standard web cookie risks apply. Mitigated by security headers (CSP, etc.). |
+| **Token in server logs** | Access log captures URL path | Log `/q/*` with short code masked or omitted. Configure Fastify logger to redact `/q/` paths. | Path still appears in server access logs (mitigated by masking). |
+| **Timing attack** | Measure response time to leak short code | Map-based lookup (`Map.get()`) — hash-based O(1), no character-by-character timing leak. No string comparison in the hot path. | None — timing side channel eliminated by design. |
+| **Token not in query params** | N/A (this is a mitigation) | Short code in URL path avoids browser history, Referer headers, and address bar exposure. | Path still appears in server access logs (mitigated by masking). |
+| **Distributed brute force** | Multiple IPs guess codes simultaneously | Global rate limit (30/min total across all IPs) in addition to per-IP limit. | Infeasible given keyspace. Global limit prevents botnet-scale attempts. |
+| **CSRF on regenerate** | Cross-origin POST to `/api/tunnel/qr/regenerate` | SameSite=lax cookies are NOT sent with cross-origin POST requests, providing CSRF protection. Endpoint requires authenticated session. | Verify SameSite=lax behavior through cloudflared tunnel. |
+
+### USENIX Security 2025 Flaw Coverage
+
+The [Zhang et al. paper](https://www.usenix.org/conference/usenixsecurity25/presentation/zhang-xin) (USENIX Security 2025, 47 of top-100 websites vulnerable, 42 CVEs) identified 6 critical design flaws. Coverage:
+
+| USENIX Flaw | Status | Implementation |
+|-------------|--------|----------------|
+| Flaw-1: Missing single-use enforcement | **Fixed** | Atomic `consumed` flag, Map-based lookup |
+| Flaw-2: Long-lived tokens | **Fixed** | 60s TTL, 90s grace, auto-rotation |
+| Flaw-3: Predictable QrId generation | **Fixed** | `crypto.randomBytes(32)` — 256-bit entropy, rejection-sampled short codes |
+| Flaw-4: Client-side QrId generation | **Fixed** | Server-side generation only |
+| Flaw-5: Missing status notification | **Fixed** | Desktop toast notification via `tunnel:qrAuthUsed` SSE event. Shows device IP/UA with [Revoke] button. |
+| Flaw-6: Inadequate session binding | **Partial** | IP + UA stored for audit. No cryptographic channel binding (requires companion app / FIDO2 — overkill for single-user). Manual revocation as active defense. |
+
+### Industry Comparison
+
+| Platform | Model | How This Plan Compares |
+|----------|-------|----------------------|
+| **Discord** | Long-lived session token, no confirmation, repeatedly exploited via QRLjacking | **Better** — single-use + TTL + notification toast |
+| **WhatsApp Web** | Pre-authenticated phone confirms "Link device?", ~60s rotation | **Comparable** rotation model; missing WhatsApp's explicit confirmation prompt (acceptable: single-user, no account selection) |
+| **Signal** | Ephemeral public key in QR, E2E encrypted channel via Signal protocol | **Below** — no cryptographic channel binding. Note: Signal's QR flow was exploited by state actors in 2025 despite stronger crypto, showing that protocol strength alone doesn't prevent social engineering. |
+| **1Password** | Noise framework E2E channel, post-quantum pre-shared keys, confirmation codes | **Below** — but 1Password is a credential manager with different threat model. Overkill for a dev tool. |
+| **FIDO2 CTAP 2.2** | BLE proximity + cryptographic binding + biometric verification | **Below** — but requires BLE stack, FIDO server, and companion authenticator. Completely inappropriate here. |
+
+### Comparison to Prior Design
+
+| Property | Original Plan | Current Plan |
+|----------|--------------|--------------|
+| Token TTL | Infinite (until restart) | 60 seconds (90s grace for previous token) |
+| Reuse | Multi-use (same QR works forever) | Single-use (consumed atomically on first scan) |
+| Secret in URL | Query param (`?t=64-char-hex`) | Opaque short code in path (`/q/Xk9mQ3`) |
+| Leak impact | Permanent access until manual revoke | Worthless after first use or 90s, whichever comes first |
+| Desktop QR refresh | Manual only | Auto-refresh every 60s via SSE with inline SVG |
+| Session binding | IP only | IP + UA stored for audit (not blocking). Manual revocation endpoint. |
+| Auth notification | None | Desktop toast: "Device [IP] authenticated via QR. Not you? [Revoke]" |
+| Audit logging | None | `session-lifecycle.jsonl` entry on every QR auth event |
+| Rate limiting | Per-IP only, shared with Basic Auth | Per-IP (separate counter) + global path limit (30/min) |
+| Short code generation | Modulo-biased | Rejection-sampled (no bias) |
+| Short code lookup | Array scan (timing leak) | Map-based O(1) (timing-safe) |
+| Connect latency | ~50ms (localhost only) | ~150-300ms through Cloudflare tunnel (honest estimate) |
+
+### What This Does NOT Protect Against
+
+- **FIDO2/passkey-level phishing resistance**: Would require BLE proximity verification and cryptographic channel binding. Overkill for a self-hosted single-user dev tool. The FIDO2 CTAP 2.2 hybrid transport is the gold standard but requires BLE hardware and a companion authenticator.
+- **Compromised phone**: If the attacker has physical access to the phone that scans, no QR scheme helps.
+- **Compromised Cloudflare tunnel**: Cloudflare terminates TLS and can inspect all traffic. This is inherent to using `trycloudflare.com` quick tunnels — use `--https` for end-to-end encryption if this matters.
+- **State-sponsored quishing**: Sophisticated attackers could create convincing phishing pages that relay the QR in real-time. The 60s rotation and desktop notification toast mitigate this for the single-user case, but a dedicated attacker with social engineering could theoretically succeed within the TTL window.
+
+### Standards Compliance Note
+
+This design is **inspired by but does not conform to** [OASIS SQRAP v1.0](https://docs.oasis-open.org/esat/sqrap/v1.0/cs01/sqrap-v1.0-cs01.html). SQRAP's architecture requires a companion mobile app with stored identity keys, public key channel binding, back-channel authentication, and user presence verification (biometric/PIN). These are fundamentally incompatible with a browser-scan-to-authenticate flow. SQRAP is referenced for awareness of formal QR auth standards, not as a compliance target.
+
+## Performance
+
+The design prioritizes speed on connect. Latency depends on whether the request goes through a Cloudflare tunnel or is localhost:
+
+### Localhost (no tunnel)
+
+| Step | Latency |
+|------|---------|
+| QR scan (physical) | ~1-2s (user action) |
+| `GET /q/:code` → Map.get() lookup + consume | <1ms |
+| Cookie set + 302 redirect | <1ms |
+| Browser follows redirect to `/` | <5ms |
+| **Total (after scan)** | **<10ms** |
+
+### Through Cloudflare Tunnel (typical mobile use case)
+
+Each request traverses: phone → Cloudflare edge (TLS termination) → cloudflared → localhost. The 302 redirect means **two full round trips** through the tunnel.
+
+| Step | Latency |
+|------|---------|
+| QR scan (physical) | ~1-2s (user action) |
+| DNS resolution for `*.trycloudflare.com` | 20-80ms (first request, cached after) |
+| TLS handshake to Cloudflare edge | 50-100ms (first request, 0 with TLS resumption) |
+| `GET /q/:code` through tunnel (request + response) | 30-90ms |
+| Browser follows 302 redirect: `GET /` through tunnel | 30-90ms |
+| **Total first connection (cold)** | **~200-400ms** |
+| **Total subsequent (TLS/DNS cached)** | **~100-200ms** |
+
+This is still fast — **imperceptible after the 1-2s physical QR scan action**. For comparison, VS Code Remote Tunnels (through Azure) adds 20-100ms per hop.
+
+### Why Not Eliminate the Redirect?
+
+The 302 means two round trips. Alternatives considered:
+- **200 + serve `index.html` directly**: URL bar shows `/q/Xk9mQ3`, relative paths break, couples auth to static serving. Not worth the complexity.
+- **200 + `<meta http-equiv="refresh">`**: Still two requests, plus HTML parse delay. Actually slower.
+- **200 + JavaScript redirect**: Same problem, plus fails if JS disabled.
+
+The 302 is clean, universally supported, and the extra 30-90ms is invisible to users.
+
+### QR Code Size Optimization
+
+The URL `https://xxx-yyy.trycloudflare.com/q/Xk9mQ3` is ~53-56 characters. At QR Error Correction Level M:
+
+| QR Version | Grid Size | Byte Capacity | Fits? |
+|------------|-----------|---------------|-------|
+| Version 3 | 29x29 | 42 bytes | No |
+| Version 4 | 33x33 | 62 bytes | Yes (comfortably) |
+| Version 5 | 37x37 | 84 bytes | Yes |
+
+The shortened `/q/` path (vs `/qr-auth/`) and 6-char code (vs 8-char) save 9 bytes, targeting Version 4 (33x33) for faster scanning on budget Android phones. Modern phones scan Version 4 QR codes in 100-300ms — the user action of pointing the camera dominates.
+
+### Desktop QR Refresh
+
+Token rotation SSE events now embed the SVG directly in the payload (~2-5KB). The desktop gets the new QR in a single SSE push — no extra HTTP fetch needed. Refresh latency: **sub-50ms** (SSE adaptive batching at 16-50ms).
+
+### SVG Caching
+
+QR SVG is cached per rotation cycle on `TunnelManager.cachedQrSvg`. The SVG is regenerated only when the token rotates (every 60s), not on every `/api/tunnel/qr` request. SVG format is optimal: resolution-independent (retina-safe), inline-able (no extra HTTP request), ~2-5KB, renders in <1ms.
+
+## Edge Cases
+
+1. **Scan during rotation**: The server keeps 2 tokens (current + previous). If the user scans right as rotation happens, the previous token is still valid for up to 60s more. Seamless.
+
+2. **Server restart**: All tokens cleared (in-memory). New token generated immediately. Tunnel URL also changes (trycloudflare gives a new subdomain), so old QR codes are doubly dead.
+
+3. **Multiple devices**: Each scan consumes the token and triggers a fresh one. To auth a second device, wait for the QR to refresh (≤60s) or hit "Regenerate QR" on the desktop, then scan the new code.
+
+4. **Token without tunnel**: `/qr-auth/:code` works even on localhost. If you have the code and it's valid, you get authenticated regardless of access method.
+
+5. **Tunnel restart (same server)**: Tokens survive tunnel restarts (stored on `TunnelManager` instance). But new tunnel URL = new QR code generated. Short code stays valid until consumed or expired.
+
+6. **Desktop browser closed during scan**: Token is consumed server-side. The scanning phone gets authenticated. When the desktop reopens, SSE reconnects and shows a fresh QR. No state corruption.
+
+7. **Race condition: two phones scan same QR**: First scanner wins (atomic `consumed = true`). Second scanner gets 401. This is correct behavior — single-use by design.
+
+## Files to Modify
+
+| File | Changes |
+|------|---------|
+| `src/tunnel-manager.ts` | `QrTokenRecord` type, `Map<shortCode, record>` token pool, rejection-sampled `generateShortCode()`, rotation timer, `consumeToken()`, `getCurrentShortCode()`, `getQrSvg()` (cached), `regenerateQrToken()`, global rate limit counter, cleanup in `stop()` |
+| `src/web/middleware/auth.ts` | Add `/q/` bypass in `onRequest` hook. Enhance session record type from `string` to `{ ip, ua, createdAt, method }` (**breaking type change** — all consumers must update). Add `qrAuthFailures` StaleExpirationMap (separate from Basic Auth `authFailures`). |
+| `src/web/routes/system-routes.ts` | Modify `/api/tunnel/qr` to use `getQrSvg()` cache. Add `GET /q/:code` with atomic consume, audit log, and `tunnel:qrAuthUsed` broadcast. Add `POST /api/tunnel/qr/regenerate`. Add `POST /api/auth/revoke`. |
+| `src/web/server.ts` | Pass `authState` + `lifecycleLog` to route context. Listen for `qrTokenRotated` and `qrTokenRegenerated` events → broadcast SSE with inline SVG. |
+| `src/web/public/app.js` | Auto-refresh QR from inline SSE SVG payload (no extra fetch). Countdown timer. Regenerate button. Auth badge. Auth notification toast on `tunnel:qrAuthUsed` with [Revoke] action. |
+| `src/session-lifecycle-log.ts` | Add `qr_auth` event type to lifecycle log schema |
+| `src/types/api.ts` | Update `AuthState` interface: `authSessions` value type, add `qrAuthFailures` map |
+
+## Complexity Estimate
+
+Medium change. Core logic (Map-based token pool, rejection-sampled short codes, SVG cache, atomic consumption, cookie issuance, audit logging) is ~120 lines. Rate limiting (separate QR counter + global path limit) adds ~20 lines. SSE plumbing with inline SVG adds ~30 lines. Frontend (inline SVG refresh, auth notification toast with revoke, countdown) is ~40 lines. Auth type migration (session record type change) touches ~10 lines across middleware. No new dependencies — `crypto` and `qrcode` are already available.
+
+## Testing
+
+### Automated
+
+```bash
+# Unit test for token manager
+npx vitest run test/qr-auth.test.ts
+```
+
+Test cases:
+- Token rotation generates unique short codes (6-char, base62)
+- Short codes have uniform character distribution (no modulo bias — verify with chi-squared test over 10K samples)
+- `consumeToken()` returns true on first use, false on second
+- Expired tokens (>90s old) return false
+- Previous token still works during 90s grace period
+- Token at exactly 60s still valid (within grace), token at 91s rejected
+- `regenerateQrToken()` invalidates all existing tokens (Map cleared)
+- Short code lookup is case-sensitive
+- Per-IP rate limiting increments on invalid codes (separate from Basic Auth counter)
+- Global rate limit (30/min) blocks attempts across all IPs
+- SVG cache returns same string for same short code, regenerates on rotation
+- Audit log entry written on successful QR auth
+- `tunnel:qrAuthUsed` SSE event broadcast on successful QR auth
+- `tunnel:qrRotated` SSE event includes inline SVG payload
+- Map-based lookup does not leak timing information (no string comparison in hot path)
+
+### Manual
+
+1. Start server with `CODEMAN_PASSWORD=test`
+2. Enable tunnel
+3. Verify `/api/tunnel/qr` returns QR encoding `https://...trycloudflare.com/q/Xk9mQ3`
+4. Open the QR URL in incognito → should auto-redirect to `/` with session cookie
+5. Verify desktop shows notification toast: "Device [IP] authenticated via QR"
+6. Open the **same** URL again → should get 401 (single-use consumed)
+7. Wait 60s → verify QR display auto-updated (new short code, inline SVG via SSE)
+8. Open just the tunnel URL → should get Basic Auth prompt
+9. Call `POST /api/tunnel/qr/regenerate` → old QR URL returns 401, new QR appears
+10. Verify per-IP rate limiting: 10+ failed `/q/badcode` → 429
+11. Verify Basic Auth failures don't consume QR rate limit budget (and vice versa)
+12. Check `~/.codeman/session-lifecycle.jsonl` for `qr_auth` entries after successful scan
+13. Click [Revoke] on the notification toast → verify session is invalidated
+
+## References
+
+- [USENIX Security 2025: "Demystifying the (In)Security of QR Code-based Login in Real-world Deployments"](https://www.usenix.org/conference/usenixsecurity25/presentation/zhang-xin) — 6 design flaws, 5 attack types, 42 CVEs across 47 of top-100 websites. Primary design reference for this plan.
+- [OWASP QRLJacking](https://owasp.org/www-community/attacks/Qrljacking) — canonical QR session hijacking reference
+- [OASIS SQRAP v1.0 Standard](https://docs.oasis-open.org/esat/sqrap/v1.0/cs01/sqrap-v1.0-cs01.html) — formal standard for secure QR authentication. **Not a compliance target** for this plan (requires companion app + PKI). Referenced for awareness only.
+- [FIDO2 CTAP 2.2 Hybrid Transport](https://fidoalliance.org/specs/fido-v2.2-rd-20230321/fido-client-to-authenticator-protocol-v2.2-rd-20230321.html) — gold standard for cross-device auth (overkill for this use case)
+- [Google GTIG: Signal QR quishing by Russian state actors (2025)](https://cloud.google.com/blog/topics/threat-intelligence/russia-targeting-signal-messenger) — UNC5792/Sandworm exploited Signal's linked-device QR flow via phishing. Demonstrates that even cryptographically strong QR auth can be defeated by social engineering.
+- [CVE-2026-2144: Magic Login QR Code Plugin race condition](https://www.cvedetails.com/cve/CVE-2026-2144/) — QR token stored as predictable static file, race window between creation and deletion. Validates this plan's in-memory-only approach.

install.sh

@@ -7,8 +7,10 @@
 # Environment variables:
 #   CODEMAN_NONINTERACTIVE=1  - Skip all prompts (for CI/automation)
 #   CODEMAN_INSTALL_DIR       - Custom install directory (default: ~/.codeman/app)
-#   CODEMAN_SKIP_SYSTEMD=1    - Skip systemd service setup prompt
+#   CODEMAN_SKIP_SYSTEMD=1    - Skip systemd/launchd service setup prompt
 #   CODEMAN_NODE_VERSION      - Node.js major version to install (default: 22)
+#   CODEMAN_REPO_URL          - Custom git repository URL (default: upstream Codeman)
+#   CODEMAN_BRANCH            - Git branch to install (default: master)
 
 set -euo pipefail
 
@@ -17,7 +19,8 @@ set -euo pipefail
 # ============================================================================
 
 INSTALL_DIR="${CODEMAN_INSTALL_DIR:-$HOME/.codeman/app}"
-REPO_URL="https://github.com/Ark0N/Codeman.git"
+REPO_URL="${CODEMAN_REPO_URL:-https://github.com/Ark0N/Codeman.git}"
+BRANCH="${CODEMAN_BRANCH:-master}"
 MIN_NODE_VERSION=18
 TARGET_NODE_VERSION="${CODEMAN_NODE_VERSION:-22}"
 NONINTERACTIVE="${CODEMAN_NONINTERACTIVE:-0}"
@@ -312,6 +315,32 @@ get_opencode_path() {
     done
 }
 
+check_cloudflared() {
+    # Check ~/.local/bin first (matches tunnel-manager.ts resolution order)
+    if [[ -x "$HOME/.local/bin/cloudflared" ]]; then
+        return 0
+    fi
+    if [[ -x "/usr/local/bin/cloudflared" ]]; then
+        return 0
+    fi
+    if command -v cloudflared &>/dev/null; then
+        return 0
+    fi
+    return 1
+}
+
+get_cloudflared_path() {
+    if [[ -x "$HOME/.local/bin/cloudflared" ]]; then
+        echo "$HOME/.local/bin/cloudflared"
+        return
+    fi
+    if [[ -x "/usr/local/bin/cloudflared" ]]; then
+        echo "/usr/local/bin/cloudflared"
+        return
+    fi
+    command -v cloudflared 2>/dev/null
+}
+
 # ============================================================================
 # Dependency Installation
 # ============================================================================
@@ -324,8 +353,15 @@ ensure_sudo() {
         die "sudo is required but not installed. Please install packages manually or run as root."
     fi
     # Validate sudo access
-    if ! sudo -v 2>/dev/null; then
-        die "Failed to obtain sudo privileges."
+    # When piped (curl | bash), stdin is the pipe — redirect from /dev/tty so sudo can prompt
+    if [[ -e /dev/tty ]]; then
+        if ! sudo -v 2>/dev/null < /dev/tty; then
+            die "Failed to obtain sudo privileges."
+        fi
+    else
+        if ! sudo -v 2>/dev/null; then
+            die "Failed to obtain sudo privileges. Try running the script directly instead of piping."
+        fi
     fi
 }
 
@@ -343,7 +379,12 @@ ensure_homebrew() {
     fi
 
     info "Installing Homebrew first..."
-    /bin/bash -c "$(download_to_stdout https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
+    # When piped (curl | bash), stdin is the pipe — Homebrew needs TTY for sudo password prompt
+    if [[ -e /dev/tty ]]; then
+        /bin/bash -c "$(download_to_stdout https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" < /dev/tty
+    else
+        NONINTERACTIVE=1 /bin/bash -c "$(download_to_stdout https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
+    fi
 
     # Add Homebrew to PATH for Apple Silicon
     if [[ -f /opt/homebrew/bin/brew ]]; then
@@ -541,6 +582,82 @@ install_git_suse() {
     run_as_root zypper install -y git
 }
 
+install_cloudflared_macos() {
+    info "Installing cloudflared via Homebrew..."
+    ensure_homebrew
+    brew install cloudflared
+}
+
+install_cloudflared_debian() {
+    info "Installing cloudflared..."
+    ensure_sudo
+    local arch
+    arch="$(dpkg --print-architecture 2>/dev/null || echo "amd64")"
+    local tmp
+    tmp="$(mktemp)"
+    download "https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-$arch.deb" "$tmp"
+    run_as_root dpkg -i "$tmp"
+    rm -f "$tmp"
+}
+
+install_cloudflared_fedora() {
+    info "Installing cloudflared..."
+    ensure_sudo
+    local arch
+    arch="$(uname -m)"
+    local rpm_arch="$arch"
+    [[ "$arch" == "x86_64" ]] && rpm_arch="x86_64"
+    [[ "$arch" == "aarch64" ]] && rpm_arch="aarch64"
+    local tmp
+    tmp="$(mktemp)"
+    download "https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-$rpm_arch.rpm" "$tmp"
+    run_as_root rpm -i "$tmp" || run_as_root rpm -U "$tmp"
+    rm -f "$tmp"
+}
+
+install_cloudflared_arch() {
+    info "Installing cloudflared binary..."
+    local arch
+    arch="$(uname -m)"
+    local cf_arch="amd64"
+    [[ "$arch" == "aarch64" ]] && cf_arch="arm64"
+    [[ "$arch" == "armv7l" ]] && cf_arch="arm"
+    ensure_sudo
+    local tmp
+    tmp="$(mktemp)"
+    download "https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-$cf_arch" "$tmp"
+    run_as_root mv "$tmp" /usr/local/bin/cloudflared
+    run_as_root chmod +x /usr/local/bin/cloudflared
+}
+
+install_cloudflared_alpine() {
+    info "Installing cloudflared binary..."
+    local arch
+    arch="$(uname -m)"
+    local cf_arch="amd64"
+    [[ "$arch" == "aarch64" ]] && cf_arch="arm64"
+    [[ "$arch" == "armv7l" ]] && cf_arch="arm"
+    ensure_sudo
+    local tmp
+    tmp="$(mktemp)"
+    download "https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-$cf_arch" "$tmp"
+    run_as_root mv "$tmp" /usr/local/bin/cloudflared
+    run_as_root chmod +x /usr/local/bin/cloudflared
+}
+
+install_cloudflared_suse() {
+    info "Installing cloudflared..."
+    ensure_sudo
+    local arch
+    arch="$(uname -m)"
+    local rpm_arch="$arch"
+    local tmp
+    tmp="$(mktemp)"
+    download "https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-$rpm_arch.rpm" "$tmp"
+    run_as_root rpm -i "$tmp" || run_as_root rpm -U "$tmp"
+    rm -f "$tmp"
+}
+
 # ============================================================================
 # Interactive Prompts
 # ============================================================================
@@ -682,9 +799,84 @@ setup_sc_alias() {
 }
 
 # ============================================================================
-# Systemd Service Setup (Linux only)
+# Service Setup (Linux systemd / macOS launchd)
 # ============================================================================
 
+setup_launchd_service() {
+    local plist_label="com.codeman.web"
+    local agent_dir="$HOME/Library/LaunchAgents"
+    local agent_plist="$agent_dir/$plist_label.plist"
+    local daemon_plist="/Library/LaunchDaemons/$plist_label.plist"
+
+    info "Setting up macOS LaunchAgent..."
+
+    # Remove any existing LaunchDaemon (system-level) to prevent duplicates.
+    # We standardize on LaunchAgent (user-level) — it doesn't require sudo,
+    # inherits the user's environment, and is the correct choice for user apps.
+    if [[ -f "$daemon_plist" ]]; then
+        warn "Found system-level LaunchDaemon at $daemon_plist — removing to prevent duplicate"
+        sudo launchctl unload "$daemon_plist" 2>/dev/null || true
+        sudo rm -f "$daemon_plist"
+        success "Removed duplicate LaunchDaemon"
+    fi
+
+    # Unload existing agent before overwriting
+    if [[ -f "$agent_plist" ]]; then
+        launchctl unload "$agent_plist" 2>/dev/null || true
+    fi
+
+    mkdir -p "$agent_dir"
+
+    # Build PATH: ensure /opt/homebrew/bin (Apple Silicon) and ~/.local/bin are included
+    local svc_path="/opt/homebrew/bin:/usr/local/bin:$HOME/.local/bin:/usr/bin:/bin:/usr/sbin:/sbin"
+
+    # Find node binary path
+    local node_path
+    node_path=$(command -v node)
+
+    cat > "$agent_plist" << EOF
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+  <key>Label</key>
+  <string>$plist_label</string>
+  <key>ProgramArguments</key>
+  <array>
+    <string>$node_path</string>
+    <string>$INSTALL_DIR/dist/index.js</string>
+    <string>web</string>
+  </array>
+  <key>EnvironmentVariables</key>
+  <dict>
+    <key>PATH</key>
+    <string>$svc_path</string>
+    <key>HOME</key>
+    <string>$HOME</string>
+    <key>LANG</key>
+    <string>en_US.UTF-8</string>
+  </dict>
+  <key>WorkingDirectory</key>
+  <string>$HOME</string>
+  <key>RunAtLoad</key>
+  <true/>
+  <key>KeepAlive</key>
+  <true/>
+  <key>ThrottleInterval</key>
+  <integer>10</integer>
+  <key>StandardOutPath</key>
+  <string>/tmp/codeman.log</string>
+  <key>StandardErrorPath</key>
+  <string>/tmp/codeman.log</string>
+</dict>
+</plist>
+EOF
+
+    launchctl load "$agent_plist" 2>/dev/null || true
+
+    success "LaunchAgent installed and started"
+}
+
 setup_systemd_service() {
     local service_dir="$HOME/.config/systemd/user"
     local service_file="$service_dir/codeman-web.service"
@@ -733,6 +925,22 @@ EOF
     success "Systemd service installed and started"
 }
 
+setup_tunnel_service() {
+    local service_dir="$HOME/.config/systemd/user"
+    local service_file="$service_dir/codeman-tunnel.service"
+
+    info "Setting up Cloudflare tunnel systemd service..."
+
+    mkdir -p "$service_dir"
+    cp "$INSTALL_DIR/scripts/codeman-tunnel.service" "$service_file"
+
+    systemctl --user daemon-reload
+    systemctl --user enable codeman-tunnel.service 2>/dev/null || true
+
+    success "Tunnel service installed (start with: systemctl --user start codeman-tunnel)"
+    echo -e "  ${DIM}Note: Set CODEMAN_PASSWORD env var before starting the tunnel for security.${NC}"
+}
+
 # ============================================================================
 # Installation Helpers
 # ============================================================================
@@ -915,6 +1123,24 @@ main() {
         fi
     fi
 
+    # cloudflared (optional — for remote/mobile access via Cloudflare Tunnel)
+    info "Checking cloudflared (optional, for remote access)..."
+    if check_cloudflared; then
+        success "cloudflared found at $(get_cloudflared_path)"
+    else
+        if prompt_yes_no "Install cloudflared? (enables remote/mobile access via Cloudflare Tunnel)" "n"; then
+            install_dependency "cloudflared" "$os" "$distro"
+            hash -r 2>/dev/null || true
+            if check_cloudflared; then
+                success "cloudflared installed at $(get_cloudflared_path)"
+            else
+                warn "cloudflared installation failed. You can install it manually later."
+            fi
+        else
+            info "Skipped (you can install cloudflared later for remote access)"
+        fi
+    fi
+
     echo ""
 
     # ========================================================================
@@ -926,26 +1152,27 @@ main() {
     if [[ -d "$INSTALL_DIR/.git" ]]; then
         info "Existing installation found, updating..."
         cd "$INSTALL_DIR"
+        git remote set-url origin "$REPO_URL" 2>/dev/null || true
 
         # Check for local changes
         if ! git diff --quiet 2>/dev/null || ! git diff --staged --quiet 2>/dev/null; then
             warn "Local changes detected in $INSTALL_DIR"
             if prompt_yes_no "Discard local changes and update?" "n"; then
                 git fetch --quiet origin
-                git reset --hard origin/master --quiet
+                git reset --hard "origin/$BRANCH" --quiet
             else
                 info "Keeping existing installation, skipping update"
             fi
         else
             git fetch --quiet origin
-            git reset --hard origin/master --quiet
+            git reset --hard "origin/$BRANCH" --quiet
         fi
     else
         # Create parent directory
         mkdir -p "$(dirname "$INSTALL_DIR")"
 
         # Clone repository (shallow for speed)
-        git clone --quiet --depth 1 "$REPO_URL" "$INSTALL_DIR"
+        git clone --quiet --depth 1 --branch "$BRANCH" "$REPO_URL" "$INSTALL_DIR"
         cd "$INSTALL_DIR"
     fi
 
@@ -989,18 +1216,7 @@ main() {
     fi
 
     # ========================================================================
-    # Systemd Service (Linux only)
-    # ========================================================================
-
-    if [[ "$os" == "linux" ]] && [[ "$SKIP_SYSTEMD" != "1" ]] && command -v systemctl &>/dev/null; then
-        echo ""
-        if prompt_yes_no "Set up systemd service for auto-start?" "n"; then
-            setup_systemd_service
-        fi
-    fi
-
-    # ========================================================================
-    # Success!
+    # Launch Options
     # ========================================================================
 
     echo ""
@@ -1009,13 +1225,85 @@ main() {
     echo -e "${GREEN}${BOLD}============================================================${NC}"
     echo ""
 
-    # Check if systemd service is running (we just started it above)
-    local service_running=false
-    if systemctl --user is-active codeman-web.service &>/dev/null; then
-        service_running=true
+    local launch_choice=""
+    local has_service=false
+    local service_type=""
+
+    if [[ "$os" == "linux" ]] && [[ "$SKIP_SYSTEMD" != "1" ]] && command -v systemctl &>/dev/null; then
+        has_service=true
+        service_type="systemd"
+    elif [[ "$os" == "macos" ]] && [[ "$SKIP_SYSTEMD" != "1" ]]; then
+        has_service=true
+        service_type="launchd"
     fi
 
-    if [[ "$service_running" == "true" ]]; then
+    if [[ "$has_service" == "true" ]]; then
+        local service_label="systemd service"
+        [[ "$service_type" == "launchd" ]] && service_label="LaunchAgent"
+
+        echo -e "  ${BOLD}How would you like to run Codeman?${NC}"
+        echo ""
+        echo -e "    ${CYAN}1)${NC} Run now in this terminal"
+        echo -e "    ${CYAN}2)${NC} Install as $service_label (auto-start on boot)"
+        echo -e "    ${CYAN}3)${NC} Don't start — I'll run it later"
+        echo ""
+
+        if [[ "$NONINTERACTIVE" == "1" ]] || [[ ! -t 0 ]]; then
+            launch_choice="3"
+        else
+            while true; do
+                echo -en "${CYAN}Choose [1/2/3]:${NC} " >&2
+                read -r launch_choice
+                case "$launch_choice" in
+                    1|2|3) break ;;
+                    *) echo "Please enter 1, 2, or 3." >&2 ;;
+                esac
+            done
+        fi
+    else
+        # No service manager available — only offer run now or skip
+        echo -e "  ${BOLD}Would you like to start Codeman now?${NC}"
+        echo ""
+        echo -e "    ${CYAN}1)${NC} Run now in this terminal"
+        echo -e "    ${CYAN}2)${NC} Don't start — I'll run it later"
+        echo ""
+
+        if [[ "$NONINTERACTIVE" == "1" ]] || [[ ! -t 0 ]]; then
+            launch_choice="2"
+        else
+            while true; do
+                echo -en "${CYAN}Choose [1/2]:${NC} " >&2
+                read -r launch_choice
+                case "$launch_choice" in
+                    1) break ;;
+                    2) break ;;
+                    *) echo "Please enter 1 or 2." >&2 ;;
+                esac
+            done
+        fi
+        # Remap: no-systemd choice "2" (skip) → internal "3"
+        [[ "$launch_choice" == "2" ]] && launch_choice="3"
+    fi
+
+    echo ""
+
+    # Handle service setup
+    if [[ "$launch_choice" == "2" ]]; then
+        if [[ "$service_type" == "launchd" ]]; then
+            setup_launchd_service
+        else
+            setup_systemd_service
+        fi
+
+        # Offer tunnel service if cloudflared is available (Linux only — systemd tunnel service)
+        if [[ "$service_type" == "systemd" ]] && check_cloudflared && [[ -f "$INSTALL_DIR/scripts/codeman-tunnel.service" ]]; then
+            echo ""
+            if prompt_yes_no "Also set up Cloudflare tunnel service? (requires CODEMAN_PASSWORD)" "n"; then
+                setup_tunnel_service
+            fi
+        fi
+
+        echo ""
         echo -e "  ${GREEN}${BOLD}Codeman is running now!${NC}"
         echo ""
         echo -e "    ${CYAN}# Open in browser${NC}"
@@ -1023,25 +1311,40 @@ main() {
         echo ""
         echo -e "  ${BOLD}Manage the service:${NC}"
         echo ""
-        echo -e "    ${CYAN}systemctl --user stop codeman-web${NC}    # Stop"
-        echo -e "    ${CYAN}systemctl --user restart codeman-web${NC} # Restart"
-        echo -e "    ${CYAN}systemctl --user status codeman-web${NC}  # Check status"
-        echo -e "    ${CYAN}journalctl --user -u codeman-web -f${NC}  # View logs"
+        if [[ "$service_type" == "launchd" ]]; then
+            echo -e "    ${CYAN}launchctl unload ~/Library/LaunchAgents/com.codeman.web.plist${NC}   # Stop"
+            echo -e "    ${CYAN}launchctl load ~/Library/LaunchAgents/com.codeman.web.plist${NC}     # Start"
+            echo -e "    ${CYAN}tail -f /tmp/codeman.log${NC}                                        # View logs"
+        else
+            echo -e "    ${CYAN}systemctl --user stop codeman-web${NC}    # Stop"
+            echo -e "    ${CYAN}systemctl --user restart codeman-web${NC} # Restart"
+            echo -e "    ${CYAN}systemctl --user status codeman-web${NC}  # Check status"
+            echo -e "    ${CYAN}journalctl --user -u codeman-web -f${NC}  # View logs"
+        fi
         echo ""
-    else
+    fi
+
+    # Show quick-start help for non-service paths
+    if [[ "$launch_choice" != "2" ]]; then
         echo -e "  ${BOLD}Quick Start:${NC}"
         echo ""
-        echo -e "    ${CYAN}# Start the web server${NC}"
-        echo -e "    codeman web"
-        echo ""
-        echo -e "    ${CYAN}# Start with HTTPS (only needed for remote access)${NC}"
-        echo -e "    codeman web --https"
+        echo -e "    ${CYAN}codeman web${NC}            # Start the web server"
+        echo -e "    ${CYAN}codeman web --https${NC}    # With HTTPS (for remote access)"
         echo ""
         echo -e "    ${CYAN}# Open in browser${NC}"
         echo -e "    http://localhost:3000"
         echo ""
     fi
 
+    if check_cloudflared; then
+        echo -e "  ${BOLD}Remote Access (Cloudflare Tunnel):${NC}"
+        echo ""
+        echo -e "    ${CYAN}./scripts/tunnel.sh start${NC}   # Start tunnel"
+        echo -e "    ${CYAN}./scripts/tunnel.sh url${NC}     # Show tunnel URL"
+        echo -e "    ${CYAN}./scripts/tunnel.sh stop${NC}    # Stop tunnel"
+        echo ""
+    fi
+
     echo -e "  ${BOLD}Mobile Access (Termius/SSH):${NC}"
     echo ""
     echo -e "    ${CYAN}sc${NC}              # Interactive tmux session chooser"
@@ -1060,16 +1363,19 @@ main() {
         echo ""
     fi
 
-    # Check if PATH needs reload in user's shell (only relevant if service not running)
-    if [[ "$service_running" != "true" ]]; then
+    # Run now in foreground (must be last — exec replaces the shell)
+    if [[ "$launch_choice" == "1" ]]; then
         local profile
         profile=$(detect_shell_profile)
-        if ! command -v codeman &>/dev/null 2>&1; then
-            echo -e "  ${YELLOW}Run this to start using codeman now:${NC}"
-            echo ""
-            echo -e "    ${CYAN}source $profile && codeman web${NC}"
-            echo ""
-        fi
+
+        echo -e "  ${GREEN}${BOLD}Starting Codeman...${NC}"
+        echo -e "  ${DIM}Press Ctrl+C to stop${NC}"
+        echo ""
+
+        # Source profile to pick up PATH changes, then exec codeman
+        # shellcheck disable=SC1090
+        source "$profile" 2>/dev/null || true
+        exec node "$INSTALL_DIR/dist/index.js" web
     fi
 }
 
@@ -1080,13 +1386,29 @@ update() {
 
     info "Updating Codeman..."
     cd "$INSTALL_DIR"
+    git remote set-url origin "$REPO_URL" 2>/dev/null || true
     git fetch --quiet origin
-    git reset --hard origin/master --quiet
+    git reset --hard "origin/$BRANCH" --quiet
     npm install --quiet --no-fund --no-audit 2>/dev/null || npm install --no-fund --no-audit
     npm run build --quiet 2>/dev/null || npm run build
     success "Updated to $(node -e "console.log(require('./package.json').version)")"
     echo ""
-    echo -e "  ${DIM}Restart codeman web to use the new version.${NC}"
+
+    # Auto-restart service if running, otherwise tell the user
+    local agent_plist="$HOME/Library/LaunchAgents/com.codeman.web.plist"
+    if systemctl --user is-active codeman-web.service &>/dev/null 2>&1; then
+        info "Restarting codeman-web service..."
+        systemctl --user restart codeman-web.service
+        success "codeman-web service restarted"
+    elif [[ -f "$agent_plist" ]]; then
+        info "Restarting LaunchAgent..."
+        launchctl unload "$agent_plist" 2>/dev/null || true
+        launchctl load "$agent_plist" 2>/dev/null || true
+        success "LaunchAgent restarted"
+    else
+        echo -e "  ${DIM}Restart codeman web to use the new version:${NC}"
+        echo -e "    ${CYAN}pkill -f 'codeman.*web'; codeman web &${NC}"
+    fi
     echo ""
 }
 
@@ -1095,20 +1417,36 @@ uninstall() {
     info "Uninstalling Codeman..."
     echo ""
 
-    # Stop and remove systemd service
-    if systemctl --user is-active codeman-web.service &>/dev/null; then
-        info "Stopping codeman-web service..."
-        systemctl --user stop codeman-web.service
+    # Stop and remove systemd services (Linux)
+    for svc in codeman-web codeman-tunnel; do
+        if systemctl --user is-active "${svc}.service" &>/dev/null 2>&1; then
+            info "Stopping ${svc} service..."
+            systemctl --user stop "${svc}.service"
+        fi
+        if systemctl --user is-enabled "${svc}.service" &>/dev/null 2>&1; then
+            info "Disabling ${svc} service..."
+            systemctl --user disable "${svc}.service" 2>/dev/null || true
+        fi
+        local svc_file="$HOME/.config/systemd/user/${svc}.service"
+        if [[ -f "$svc_file" ]]; then
+            rm -f "$svc_file"
+            success "Removed ${svc} service"
+        fi
+    done
+    systemctl --user daemon-reload 2>/dev/null || true
+
+    # Stop and remove launchd services (macOS)
+    local agent_plist="$HOME/Library/LaunchAgents/com.codeman.web.plist"
+    local daemon_plist="/Library/LaunchDaemons/com.codeman.web.plist"
+    if [[ -f "$agent_plist" ]]; then
+        launchctl unload "$agent_plist" 2>/dev/null || true
+        rm -f "$agent_plist"
+        success "Removed LaunchAgent"
     fi
-    if systemctl --user is-enabled codeman-web.service &>/dev/null 2>&1; then
-        info "Disabling codeman-web service..."
-        systemctl --user disable codeman-web.service 2>/dev/null || true
-    fi
-    local service_file="$HOME/.config/systemd/user/codeman-web.service"
-    if [[ -f "$service_file" ]]; then
-        rm -f "$service_file"
-        systemctl --user daemon-reload 2>/dev/null || true
-        success "Systemd service removed"
+    if [[ -f "$daemon_plist" ]]; then
+        sudo launchctl unload "$daemon_plist" 2>/dev/null || true
+        sudo rm -f "$daemon_plist"
+        success "Removed LaunchDaemon"
     fi
 
     # Remove symlinks
@@ -1156,5 +1494,12 @@ uninstall() {
 case "${1:-}" in
     update)    update ;;
     uninstall) uninstall ;;
-    *)         main "$@" ;;
+    *)
+        if [[ -z "${1:-}" && -d "$INSTALL_DIR/.git" ]]; then
+            print_banner
+            update
+        else
+            main "$@"
+        fi
+        ;;
 esac

package.json

@@ -1,6 +1,6 @@
 {
   "name": "aicodeman",
-  "version": "0.2.9",
+  "version": "0.5.11",
   "description": "The missing control plane for AI coding agents - run 20 autonomous agents with real-time monitoring and session persistence",
   "type": "module",
   "main": "dist/index.js",
@@ -11,16 +11,16 @@
   "scripts": {
     "postinstall": "node scripts/postinstall.js",
     "build": "node scripts/build.mjs",
-    "start": "node dist/index.js",
+    "start": "NODE_COMPILE_CACHE=${HOME}/.codeman/compile-cache node dist/index.js",
     "dev": "tsx src/index.ts web",
     "web": "node dist/index.js web",
     "clean": "rm -rf dist",
-    "test": "vitest run",
-    "test:watch": "vitest",
-    "test:coverage": "vitest run --coverage",
+    "test": "vitest run --config config/vitest.config.ts",
+    "test:watch": "vitest --config config/vitest.config.ts",
+    "test:coverage": "vitest run --config config/vitest.config.ts --coverage",
     "typecheck": "tsc --noEmit",
-    "lint": "eslint 'src/**/*.ts'",
-    "lint:fix": "eslint 'src/**/*.ts' --fix",
+    "lint": "eslint --config config/eslint.config.js 'src/**/*.ts'",
+    "lint:fix": "eslint --config config/eslint.config.js 'src/**/*.ts' --fix",
     "format": "prettier --write 'src/**/*.ts'",
     "format:check": "prettier --check 'src/**/*.ts'",
     "capture:subagents": "node scripts/capture-subagent-screenshots.mjs",
@@ -51,6 +51,11 @@
     "@fastify/compress": "^8.3.1",
     "@fastify/cookie": "^11.0.2",
     "@fastify/static": "^8.0.0",
+    "@fastify/websocket": "^11.2.0",
+    "@xterm/addon-fit": "^0.11.0",
+    "@xterm/addon-unicode11": "^0.9.0",
+    "@xterm/addon-webgl": "^0.19.0",
+    "@xterm/xterm": "^6.0.0",
     "chalk": "^5.3.0",
     "chokidar": "^3.6.0",
     "commander": "^12.1.0",
@@ -59,10 +64,6 @@
     "qrcode": "^1.5.4",
     "uuid": "^10.0.0",
     "web-push": "^3.6.7",
-    "xterm": "^5.3.0",
-    "xterm-addon-fit": "^0.8.0",
-    "xterm-addon-unicode11": "^0.6.0",
-    "xterm-addon-webgl": "^0.16.0",
     "zod": "^4.3.6"
   },
   "devDependencies": {
@@ -72,9 +73,11 @@
     "@remotion/transitions": "4.0.429",
     "@types/node": "^20.19.33",
     "@types/pngjs": "^6.0.5",
+    "@types/qrcode": "^1.5.6",
     "@types/react": "^19.2.14",
     "@types/uuid": "^10.0.0",
     "@types/web-push": "^3.6.4",
+    "@types/ws": "^8.18.1",
     "@vitest/coverage-v8": "^4.0.18",
     "agent-browser": "^0.6.0",
     "esbuild": "^0.27.3",
@@ -90,6 +93,10 @@
     "typescript-eslint": "^8.0.0",
     "vitest": "^4.0.18"
   },
+  "optionalDependencies": {
+    "@remotion/compositor-linux-x64-gnu": "^4.0.432",
+    "@rspack/binding-linux-x64-gnu": "^1.7.7"
+  },
   "engines": {
     "node": ">=18.0.0"
   },

packages/xterm-zerolag-input/package.json

@@ -17,7 +17,7 @@
     "dist/"
   ],
   "scripts": {
-    "build": "tsup src/index.ts --format cjs,esm --dts --clean",
+    "build": "tsup",
     "test": "vitest run",
     "typecheck": "tsc --noEmit",
     "prepublishOnly": "npm run build"

packages/xterm-zerolag-input/src/cell-dimensions.ts

@@ -4,18 +4,26 @@ import type { XtermTerminal, CellDimensions } from './types.js';
  * Get cell dimensions from the terminal, handling xterm.js v5 (private API)
  * and v7+ (public API).
  *
+ * Returns CSS-pixel values.  xterm's `device.char` is in device pixels, so
+ * we divide by `devicePixelRatio` to stay consistent with `css.cell`.
+ *
  * Returns `null` if the terminal is not yet rendered or dimensions are
  * unavailable.
  */
 export function getCellDimensions(terminal: XtermTerminal): CellDimensions | null {
     // eslint-disable-next-line @typescript-eslint/no-explicit-any
     const t = terminal as any;
+    const dpr = typeof devicePixelRatio === 'number' && devicePixelRatio > 0
+        ? devicePixelRatio : 1;
 
     // Try v7+ public API first
     if (t.dimensions?.css?.cell) {
+        const cellH = t.dimensions.css.cell.height;
         return {
             width: t.dimensions.css.cell.width,
-            height: t.dimensions.css.cell.height,
+            height: cellH,
+            charTop: (t.dimensions?.device?.char?.top ?? 0) / dpr,
+            charHeight: (t.dimensions?.device?.char?.height ?? (cellH * dpr)) / dpr,
         };
     }
 
@@ -23,9 +31,12 @@ export function getCellDimensions(terminal: XtermTerminal): CellDimensions | nul
     try {
         const dims = t._core?._renderService?.dimensions;
         if (dims?.css?.cell) {
+            const cellH = dims.css.cell.height;
             return {
                 width: dims.css.cell.width,
-                height: dims.css.cell.height,
+                height: cellH,
+                charTop: (dims.device?.char?.top ?? 0) / dpr,
+                charHeight: (dims.device?.char?.height ?? (cellH * dpr)) / dpr,
             };
         }
     } catch {

packages/xterm-zerolag-input/src/overlay-renderer.ts

@@ -1,4 +1,50 @@
-import type { RenderParams, FontStyle } from './types.js';
+import type { RenderParams, FontStyle, XtermTerminal } from './types.js';
+
+// ─── CJK / fullwidth character width detection ───────────────────────
+
+/**
+ * Get visual cell width of a single character.
+ * CJK wide characters occupy 2 cells, others occupy 1.
+ * Prefers the terminal's Unicode addon when available.
+ */
+export function charCellWidth(terminal: XtermTerminal | null | undefined, ch: string): number {
+  if (terminal?.unicode?.getStringCellWidth) {
+    return terminal.unicode.getStringCellWidth(ch);
+  }
+  // Fallback: detect CJK wide characters by Unicode range
+  const code = ch.codePointAt(0);
+  if (
+    code !== undefined &&
+    code >= 0x1100 &&
+    (code <= 0x115f || // Hangul Jamo
+      (code >= 0x2e80 && code <= 0x303e) || // CJK Radicals, Kangxi, Ideographic
+      (code >= 0x3040 && code <= 0x33bf) || // Hiragana, Katakana, Bopomofo, CJK Compat
+      (code >= 0x3400 && code <= 0x4dbf) || // CJK Unified Ext A
+      (code >= 0x4e00 && code <= 0xa4cf) || // CJK Unified, Yi
+      (code >= 0xa960 && code <= 0xa97c) || // Hangul Jamo Extended-A
+      (code >= 0xac00 && code <= 0xd7a3) || // Hangul Syllables
+      (code >= 0xf900 && code <= 0xfaff) || // CJK Compat Ideographs
+      (code >= 0xfe30 && code <= 0xfe6f) || // CJK Compat Forms
+      (code >= 0xff01 && code <= 0xff60) || // Fullwidth Forms
+      (code >= 0xffe0 && code <= 0xffe6) || // Fullwidth Signs
+      (code >= 0x1f000 && code <= 0x1fbff) || // Mahjong, Domino, Emoji
+      (code >= 0x20000 && code <= 0x2ffff) || // CJK Unified Ext B-F
+      (code >= 0x30000 && code <= 0x3ffff)) // CJK Unified Ext G+
+  )
+    return 2;
+  return 1;
+}
+
+/**
+ * Get visual cell width of a string (sum of all character widths).
+ */
+export function stringCellWidth(terminal: XtermTerminal | null | undefined, str: string): number {
+  let w = 0;
+  for (const ch of str) w += charCellWidth(terminal, ch);
+  return w;
+}
+
+// ─── Overlay rendering ────────────────────────────────────────────────
 
 /**
  * Render the overlay content into the container element.
@@ -6,91 +52,114 @@ import type { RenderParams, FontStyle } from './types.js';
  * Creates per-character `<span>` elements positioned on an exact grid
  * matching xterm.js's canvas renderer. This avoids sub-pixel drift that
  * occurs with normal DOM text flow.
+ *
+ * CJK wide characters are rendered with double-width spans.
  */
 export function renderOverlay(container: HTMLDivElement, params: RenderParams): void {
-    const { lines, startCol, totalCols, cellW, cellH, promptRow, font, showCursor, cursorColor } = params;
+  const {
+    lines,
+    startCol,
+    totalCols,
+    cellW,
+    cellH,
+    charTop,
+    charHeight,
+    promptRow,
+    font,
+    showCursor,
+    cursorColor,
+    terminal,
+  } = params;
 
-    // Position container at prompt row
-    container.style.left = '0px';
-    container.style.top = (promptRow * cellH) + 'px';
+  // Position container at prompt row.
+  container.style.left = '0px';
+  container.style.top = promptRow * cellH + 'px';
 
-    // Clear and rebuild (typically 1-3 line divs, negligible cost)
-    container.innerHTML = '';
-    const fullWidthPx = totalCols * cellW;
+  // Clear and rebuild (typically 1-3 line divs, negligible cost)
+  container.innerHTML = '';
+  const fullWidthPx = totalCols * cellW;
 
-    for (let i = 0; i < lines.length; i++) {
-        const leftPx = i === 0 ? startCol * cellW : 0;
-        const widthPx = i === 0 ? (fullWidthPx - leftPx) : fullWidthPx;
-        const topPx = i * cellH;
-        const lineEl = makeLine(lines[i], leftPx, topPx, widthPx, cellH, cellW, font);
-        container.appendChild(lineEl);
+  for (let i = 0; i < lines.length; i++) {
+    const leftPx = i === 0 ? startCol * cellW : 0;
+    const widthPx = i === 0 ? fullWidthPx - leftPx : fullWidthPx;
+    const topPx = i * cellH;
+    const lineEl = makeLine(lines[i], leftPx, topPx, widthPx, cellH, cellW, charTop, charHeight, font, terminal);
+    container.appendChild(lineEl);
+  }
+
+  // Block cursor at end of last line (use visual width for CJK support)
+  if (showCursor) {
+    const lastLine = lines[lines.length - 1];
+    const lastLineLeft = lines.length === 1 ? startCol : 0;
+    const cursorCol = lastLineLeft + stringCellWidth(terminal, lastLine);
+    if (cursorCol < totalCols) {
+      const cursor = document.createElement('span');
+      cursor.style.cssText = 'position:absolute;display:inline-block';
+      cursor.style.left = cursorCol * cellW + 'px';
+      cursor.style.top = (lines.length - 1) * cellH + 'px';
+      cursor.style.width = cellW + 'px';
+      cursor.style.height = cellH + 'px';
+      cursor.style.backgroundColor = cursorColor;
+      container.appendChild(cursor);
     }
+  }
 
-    // Block cursor at end of last line
-    if (showCursor) {
-        const lastLine = lines[lines.length - 1];
-        const lastLineLeft = lines.length === 1 ? startCol : 0;
-        const cursorCol = lastLineLeft + lastLine.length;
-        if (cursorCol < totalCols) {
-            const cursor = document.createElement('span');
-            cursor.style.cssText = 'position:absolute;display:inline-block';
-            cursor.style.left = (cursorCol * cellW) + 'px';
-            cursor.style.top = ((lines.length - 1) * cellH) + 'px';
-            cursor.style.width = cellW + 'px';
-            cursor.style.height = cellH + 'px';
-            cursor.style.backgroundColor = cursorColor;
-            container.appendChild(cursor);
-        }
-    }
-
-    container.style.display = '';
+  container.style.display = '';
 }
 
 /**
  * Create a styled line `<div>` with per-character grid positioning.
  *
- * Each character gets its own `<span>` placed at `i * cellW` pixels.
- * This matches xterm's canvas renderer where each glyph occupies exactly
- * one cell width, regardless of the actual glyph metrics.
+ * Each character gets its own `<span>` positioned by visual column offset.
+ * CJK wide characters occupy 2 cell widths.
  */
 function makeLine(
-    text: string,
-    leftPx: number,
-    topPx: number,
-    widthPx: number,
-    cellH: number,
-    cellW: number,
-    font: FontStyle,
+  text: string,
+  leftPx: number,
+  topPx: number,
+  widthPx: number,
+  cellH: number,
+  cellW: number,
+  _charTop: number,
+  _charHeight: number,
+  font: FontStyle,
+  terminal?: XtermTerminal | null
 ): HTMLDivElement {
-    const el = document.createElement('div');
-    el.style.cssText = 'position:absolute;pointer-events:none';
-    el.style.backgroundColor = font.backgroundColor;
-    el.style.left = leftPx + 'px';
-    el.style.top = topPx + 'px';
-    el.style.width = widthPx + 'px';
-    el.style.height = (cellH + 1) + 'px';
-    el.style.lineHeight = cellH + 'px';
+  const el = document.createElement('div');
+  el.style.cssText = 'position:absolute;pointer-events:none';
+  el.style.backgroundColor = font.backgroundColor;
+  el.style.left = leftPx + 'px';
+  el.style.top = topPx + 'px';
+  el.style.width = widthPx + 'px';
+  // Extend background 1px past cell boundary to cover the compositing
+  // seam between the overlay layer (z-index:7) and the canvas layer below.
+  // The extra 1px lands in the next row's charTop gap (empty area before
+  // text rendering starts), so no canvas content is obscured.
+  el.style.height = cellH + 1 + 'px';
 
-    for (let i = 0; i < text.length; i++) {
-        const span = document.createElement('span');
-        // Match xterm.js canvas text rendering:
-        // - antialiased smoothing (canvas uses grayscale, not LCD subpixel)
-        // - geometricPrecision for consistent glyph sizing
-        // - no ligatures (canvas renders each glyph independently)
-        span.style.cssText =
-            'position:absolute;display:inline-block;text-align:center;pointer-events:none;' +
-            '-webkit-font-smoothing:antialiased;-moz-osx-font-smoothing:grayscale;' +
-            "text-rendering:geometricPrecision;font-feature-settings:'liga' 0,'calt' 0";
-        span.style.left = (i * cellW) + 'px';
-        span.style.width = cellW + 'px';
-        span.style.fontFamily = font.fontFamily;
-        span.style.fontSize = font.fontSize;
-        span.style.fontWeight = font.fontWeight;
-        span.style.color = font.color;
-        if (font.letterSpacing) span.style.letterSpacing = font.letterSpacing;
-        span.textContent = text[i];
-        el.appendChild(span);
-    }
+  // CJK wide chars occupy 2 cells — position by visual column offset
+  let colOffset = 0;
+  for (const ch of text) {
+    const cw = charCellWidth(terminal, ch);
+    const span = document.createElement('span');
+    // No ligatures — canvas renders each glyph independently.
+    span.style.cssText =
+      'position:absolute;display:inline-block;text-align:center;pointer-events:none;' +
+      "font-feature-settings:'liga' 0,'calt' 0";
+    span.style.left = colOffset * cellW + 'px';
+    span.style.top = '0px';
+    span.style.width = cw * cellW + 'px';
+    span.style.height = cellH + 'px';
+    span.style.lineHeight = cellH + 'px';
+    span.style.fontFamily = font.fontFamily;
+    span.style.fontSize = font.fontSize;
+    span.style.fontWeight = font.fontWeight;
+    span.style.color = font.color;
+    if (font.letterSpacing) span.style.letterSpacing = font.letterSpacing;
+    span.textContent = ch;
+    el.appendChild(span);
+    colOffset += cw;
+  }
 
-    return el;
+  return el;
 }

packages/xterm-zerolag-input/src/types.ts

@@ -5,28 +5,35 @@
  * Consumers pass their real Terminal instance — we only use these properties.
  */
 export interface XtermTerminal {
-    readonly element: HTMLElement | undefined;
-    readonly cols: number;
-    readonly rows: number;
-    readonly options: {
-        fontFamily?: string;
-        fontSize?: number;
-        fontWeight?: string | number;
-        theme?: {
-            background?: string;
-            foreground?: string;
-            cursor?: string;
-        };
+  readonly element: HTMLElement | undefined;
+  readonly cols: number;
+  readonly rows: number;
+  readonly options: {
+    fontFamily?: string;
+    fontSize?: number;
+    fontWeight?: string | number;
+    theme?: {
+      background?: string;
+      foreground?: string;
+      cursor?: string;
     };
-    readonly buffer: {
-        readonly active: {
-            readonly viewportY: number;
-            readonly baseY: number;
-            getLine(y: number): {
-                translateToString(trimRight?: boolean): string;
-            } | undefined;
-        };
+  };
+  readonly buffer: {
+    readonly active: {
+      readonly viewportY: number;
+      readonly baseY: number;
+      getLine(y: number):
+        | {
+            translateToString(trimRight?: boolean): string;
+          }
+        | undefined;
     };
+  };
+  /** Unicode addon (e.g. Unicode11Addon) for CJK wide character width */
+  readonly unicode?: {
+    getStringCellWidth(str: string): number;
+    activeVersion?: string;
+  };
 }
 
 /**
@@ -35,18 +42,18 @@ export interface XtermTerminal {
  * The consumer calls `terminal.loadAddon(addon)` which invokes `activate()`.
  */
 export interface XtermAddon {
-    activate(terminal: XtermTerminal): void;
-    dispose(): void;
+  activate(terminal: XtermTerminal): void;
+  dispose(): void;
 }
 
 /**
  * Position of the prompt in the terminal viewport.
  */
 export interface PromptPosition {
-    /** Viewport-relative row (0 = top of viewport) */
-    row: number;
-    /** Column of the prompt marker character */
-    col: number;
+  /** Viewport-relative row (0 = top of viewport) */
+  row: number;
+  /** Column of the prompt marker character */
+  col: number;
 }
 
 /**
@@ -60,104 +67,114 @@ export interface PromptPosition {
  * - `custom`: Full escape hatch — provide your own finder function
  */
 export type PromptFinder =
-    | { type: 'character'; char: string; offset?: number }
-    | { type: 'regex'; pattern: RegExp; offset?: number }
-    | { type: 'custom'; find: (terminal: XtermTerminal) => PromptPosition | null; offset?: number };
+  | { type: 'character'; char: string; offset?: number }
+  | { type: 'regex'; pattern: RegExp; offset?: number }
+  | { type: 'custom'; find: (terminal: XtermTerminal) => PromptPosition | null; offset?: number };
 
 /**
  * Configuration options for ZerolagInputAddon.
  */
 export interface ZerolagInputOptions {
-    /**
-     * How to find the prompt in the terminal buffer.
-     *
-     * The `offset` controls how many characters after the prompt marker
-     * the user input begins (e.g., `"> "` = offset 2).
-     *
-     * @default { type: 'character', char: '>', offset: 2 }
-     */
-    prompt?: PromptFinder;
+  /**
+   * How to find the prompt in the terminal buffer.
+   *
+   * The `offset` controls how many characters after the prompt marker
+   * the user input begins (e.g., `"> "` = offset 2).
+   *
+   * @default { type: 'character', char: '>', offset: 2 }
+   */
+  prompt?: PromptFinder;
 
-    /**
-     * Z-index for the overlay element.
-     * @default 7
-     */
-    zIndex?: number;
+  /**
+   * Z-index for the overlay element.
+   * @default 7
+   */
+  zIndex?: number;
 
-    /**
-     * Background color for the overlay.
-     * Set to `'transparent'` to disable the opaque background.
-     * @default Read from terminal.options.theme.background
-     */
-    backgroundColor?: string;
+  /**
+   * Background color for the overlay.
+   * Set to `'transparent'` to disable the opaque background.
+   * @default Read from terminal.options.theme.background
+   */
+  backgroundColor?: string;
 
-    /**
-     * Foreground color for overlay text.
-     * @default Read from terminal.options.theme.foreground
-     */
-    foregroundColor?: string;
+  /**
+   * Foreground color for overlay text.
+   * @default Read from terminal.options.theme.foreground
+   */
+  foregroundColor?: string;
 
-    /**
-     * Whether to show a block cursor at the end of the overlay text.
-     * @default true
-     */
-    showCursor?: boolean;
+  /**
+   * Whether to show a block cursor at the end of the overlay text.
+   * @default true
+   */
+  showCursor?: boolean;
 
-    /**
-     * Cursor color (block cursor at end of text).
-     * @default Read from terminal.options.theme.cursor
-     */
-    cursorColor?: string;
+  /**
+   * Cursor color (block cursor at end of text).
+   * @default Read from terminal.options.theme.cursor
+   */
+  cursorColor?: string;
 
-    /**
-     * Scroll debounce time in ms for re-rendering when user scrolls
-     * back to the bottom of the terminal.
-     * @default 50
-     */
-    scrollDebounceMs?: number;
+  /**
+   * Scroll debounce time in ms for re-rendering when user scrolls
+   * back to the bottom of the terminal.
+   * @default 50
+   */
+  scrollDebounceMs?: number;
 }
 
 /**
  * Read-only state snapshot of the overlay.
  */
 export interface ZerolagInputState {
-    /** Characters typed but not yet acknowledged by the server */
-    pendingText: string;
-    /** Number of characters flushed to PTY but echo not yet received */
-    flushedLength: number;
-    /** Text content of the flushed portion */
-    flushedText: string;
-    /** Whether the overlay is currently visible */
-    visible: boolean;
-    /** Last detected prompt position, if any */
-    promptPosition: PromptPosition | null;
+  /** Characters typed but not yet acknowledged by the server */
+  pendingText: string;
+  /** Number of characters flushed to PTY but echo not yet received */
+  flushedLength: number;
+  /** Text content of the flushed portion */
+  flushedText: string;
+  /** Whether the overlay is currently visible */
+  visible: boolean;
+  /** Last detected prompt position, if any */
+  promptPosition: PromptPosition | null;
 }
 
 /** Cell dimensions in CSS pixels. */
 export interface CellDimensions {
-    width: number;
-    height: number;
+  width: number;
+  height: number;
+  /** Vertical offset (px) from cell top to where characters render. */
+  charTop: number;
+  /** Height of the character rendering area (px). */
+  charHeight: number;
 }
 
 /** Parameters for the overlay renderer. */
 export interface RenderParams {
-    lines: string[];
-    startCol: number;
-    totalCols: number;
-    cellW: number;
-    cellH: number;
-    promptRow: number;
-    font: FontStyle;
-    showCursor: boolean;
-    cursorColor: string;
+  lines: string[];
+  startCol: number;
+  totalCols: number;
+  cellW: number;
+  cellH: number;
+  /** Vertical offset (px) from cell top to character rendering area. */
+  charTop: number;
+  /** Height of the character rendering area (px). */
+  charHeight: number;
+  promptRow: number;
+  font: FontStyle;
+  showCursor: boolean;
+  cursorColor: string;
+  /** Terminal instance for CJK wide character width detection */
+  terminal?: XtermTerminal | null;
 }
 
 /** Cached font style properties for overlay rendering. */
 export interface FontStyle {
-    fontFamily: string;
-    fontSize: string;
-    fontWeight: string;
-    color: string;
-    backgroundColor: string;
-    letterSpacing: string;
+  fontFamily: string;
+  fontSize: string;
+  fontWeight: string;
+  color: string;
+  backgroundColor: string;
+  letterSpacing: string;
 }

packages/xterm-zerolag-input/test/cell-dimensions.test.ts

@@ -0,0 +1,127 @@
+import { describe, it, expect, afterEach, beforeEach } from 'vitest';
+import { getCellDimensions } from '../src/cell-dimensions.js';
+import { createMockTerminal } from './helpers.js';
+import type { XtermTerminal } from '../src/types.js';
+
+let cleanups: (() => void)[] = [];
+
+afterEach(() => {
+    for (const fn of cleanups) fn();
+    cleanups = [];
+});
+
+describe('getCellDimensions', () => {
+    describe('v5 private API (mock _core._renderService)', () => {
+        it('returns cell width and height from css.cell', () => {
+            const mock = createMockTerminal({ cellWidth: 8.4, cellHeight: 19 });
+            cleanups.push(mock.cleanup);
+            const dims = getCellDimensions(mock.terminal as unknown as XtermTerminal);
+            expect(dims).not.toBeNull();
+            expect(dims!.width).toBe(8.4);
+            expect(dims!.height).toBe(19);
+        });
+
+        it('returns charTop from device.char.top divided by DPR', () => {
+            const mock = createMockTerminal({
+                cellWidth: 8, cellHeight: 19,
+                deviceCharTop: 2,
+            });
+            cleanups.push(mock.cleanup);
+            const dims = getCellDimensions(mock.terminal as unknown as XtermTerminal);
+            expect(dims).not.toBeNull();
+            // DPR=1 in jsdom, so charTop = 2 / 1 = 2
+            expect(dims!.charTop).toBe(2);
+        });
+
+        it('returns charHeight from device.char.height divided by DPR', () => {
+            const mock = createMockTerminal({
+                cellWidth: 8, cellHeight: 19,
+                deviceCharHeight: 16,
+            });
+            cleanups.push(mock.cleanup);
+            const dims = getCellDimensions(mock.terminal as unknown as XtermTerminal);
+            expect(dims).not.toBeNull();
+            // DPR=1, so charHeight = 16 / 1 = 16
+            expect(dims!.charHeight).toBe(16);
+        });
+
+        it('defaults charTop to 0 when device.char not present', () => {
+            // Default mock has deviceCharTop=0
+            const mock = createMockTerminal({ cellWidth: 8, cellHeight: 19 });
+            cleanups.push(mock.cleanup);
+            const dims = getCellDimensions(mock.terminal as unknown as XtermTerminal);
+            expect(dims!.charTop).toBe(0);
+        });
+
+        it('defaults charHeight to cellH when device.char.height not set', () => {
+            // Default mock has deviceCharHeight=cellH
+            const mock = createMockTerminal({ cellWidth: 8, cellHeight: 19 });
+            cleanups.push(mock.cleanup);
+            const dims = getCellDimensions(mock.terminal as unknown as XtermTerminal);
+            expect(dims!.charHeight).toBe(19);
+        });
+    });
+
+    describe('DPR simulation', () => {
+        const originalDPR = globalThis.devicePixelRatio;
+
+        beforeEach(() => {
+            // Set DPR=2 to test division
+            Object.defineProperty(globalThis, 'devicePixelRatio', {
+                value: 2,
+                writable: true,
+                configurable: true,
+            });
+        });
+
+        afterEach(() => {
+            Object.defineProperty(globalThis, 'devicePixelRatio', {
+                value: originalDPR,
+                writable: true,
+                configurable: true,
+            });
+        });
+
+        it('divides device.char.top by DPR', () => {
+            const mock = createMockTerminal({
+                cellWidth: 16, cellHeight: 38,
+                deviceCharTop: 4,
+                deviceCharHeight: 32,
+            });
+            cleanups.push(mock.cleanup);
+            const dims = getCellDimensions(mock.terminal as unknown as XtermTerminal);
+            expect(dims).not.toBeNull();
+            // charTop = 4 / 2 = 2
+            expect(dims!.charTop).toBe(2);
+            // charHeight = 32 / 2 = 16
+            expect(dims!.charHeight).toBe(16);
+        });
+    });
+
+    describe('null cases', () => {
+        it('returns null for terminal without _core', () => {
+            const terminal = {
+                element: document.createElement('div'),
+                cols: 80,
+                rows: 24,
+                options: {},
+                buffer: { active: { viewportY: 0, baseY: 0, getLine: () => undefined } },
+            } as unknown as XtermTerminal;
+            const dims = getCellDimensions(terminal);
+            expect(dims).toBeNull();
+        });
+
+        it('returns null for terminal with no dimensions', () => {
+            const terminal = {
+                element: document.createElement('div'),
+                cols: 80,
+                rows: 24,
+                options: {},
+                buffer: { active: { viewportY: 0, baseY: 0, getLine: () => undefined } },
+                _core: { _renderService: {} },
+            } as unknown as XtermTerminal;
+            const dims = getCellDimensions(terminal);
+            expect(dims).toBeNull();
+        });
+    });
+});

packages/xterm-zerolag-input/test/cjk-testplan.test.ts

@@ -0,0 +1,843 @@
+/**
+ * Comprehensive CJK wide character test plan for PR #30.
+ *
+ * Tests all 7 items from the test plan:
+ * 1. Chinese text input — no character overlap
+ * 2. CJK text renders with correct double-width spacing
+ * 3. Japanese (こんにちは) and Korean (안녕하세요) input
+ * 4. Cursor positions correctly after CJK characters
+ * 5. Long CJK input wraps at correct column boundary
+ * 6. Existing ASCII input is unaffected
+ * 7. Teammate terminal panels render CJK correctly (app.js embedded copy)
+ */
+
+import { describe, it, expect, afterEach } from 'vitest';
+import { renderOverlay, charCellWidth, stringCellWidth } from '../src/overlay-renderer.js';
+import type { RenderParams, FontStyle } from '../src/types.js';
+import { createMockTerminal } from './helpers.js';
+import { ZerolagInputAddon } from '../src/zerolag-input-addon.js';
+
+// ─── Shared fixtures ─────────────────────────────────────────────────
+
+const FONT: FontStyle = {
+  fontFamily: 'monospace',
+  fontSize: '14px',
+  fontWeight: 'normal',
+  color: '#eeeeee',
+  backgroundColor: '#0d0d0d',
+  letterSpacing: '',
+};
+
+function makeParams(overrides: Partial<RenderParams> = {}): RenderParams {
+  return {
+    lines: ['hello'],
+    startCol: 2,
+    totalCols: 80,
+    cellW: 10,
+    cellH: 17,
+    charTop: 2,
+    charHeight: 14,
+    promptRow: 0,
+    font: FONT,
+    showCursor: true,
+    cursorColor: '#e0e0e0',
+    ...overrides,
+  };
+}
+
+/** Extract span data from a rendered line div */
+function getSpans(lineDiv: HTMLDivElement) {
+  const spans: { text: string; left: number; width: number }[] = [];
+  for (let i = 0; i < lineDiv.children.length; i++) {
+    const span = lineDiv.children[i] as HTMLSpanElement;
+    spans.push({
+      text: span.textContent || '',
+      left: parseFloat(span.style.left),
+      width: parseFloat(span.style.width),
+    });
+  }
+  return spans;
+}
+
+// ─── Addon setup helpers ─────────────────────────────────────────────
+
+let cleanups: (() => void)[] = [];
+
+afterEach(() => {
+  for (const fn of cleanups) fn();
+  cleanups = [];
+});
+
+function tracked(lines: string[] = ['$ '], promptChar = '$') {
+  const mock = createMockTerminal({ buffer: { lines }, cols: 80 });
+  const addon = new ZerolagInputAddon({
+    prompt: { type: 'character', char: promptChar, offset: 2 },
+  });
+  mock.terminal.loadAddon(addon);
+  cleanups.push(() => {
+    addon.dispose();
+    mock.cleanup();
+  });
+  return { addon, mock };
+}
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 1: Chinese text input — no character overlap
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 1: Chinese text — no character overlap', () => {
+  it('consecutive Chinese characters have contiguous non-overlapping spans', () => {
+    const container = document.createElement('div');
+    const text = '你好世界';
+    renderOverlay(container, makeParams({ lines: [text], cellW: 10 }));
+
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const spans = getSpans(lineDiv);
+
+    expect(spans.length).toBe(4);
+
+    // Each span: left = previous span's (left + width), width = 20px (2 cells)
+    for (let i = 0; i < spans.length; i++) {
+      expect(spans[i].width).toBe(20); // 2 cells * 10px
+      if (i > 0) {
+        const expectedLeft = spans[i - 1].left + spans[i - 1].width;
+        expect(spans[i].left).toBe(expectedLeft);
+      }
+    }
+  });
+
+  it('Chinese sentence (simulated pinyin output) renders without gaps or overlaps', () => {
+    const container = document.createElement('div');
+    const text = '我是一个测试';
+    renderOverlay(container, makeParams({ lines: [text], cellW: 8 }));
+
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const spans = getSpans(lineDiv);
+
+    expect(spans.length).toBe(6);
+
+    // Verify contiguous positioning
+    let expectedLeft = 0;
+    for (const span of spans) {
+      expect(span.left).toBe(expectedLeft);
+      expect(span.width).toBe(16); // 2 * 8px
+      expectedLeft += span.width;
+    }
+
+    // Total visual width should be 6 chars * 2 cells * 8px = 96px
+    expect(expectedLeft).toBe(96);
+  });
+
+  it('mixed Chinese + ASCII has no gaps between spans', () => {
+    const container = document.createElement('div');
+    const text = 'hello你好world';
+    renderOverlay(container, makeParams({ lines: [text], cellW: 10 }));
+
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const spans = getSpans(lineDiv);
+
+    // h(10) e(10) l(10) l(10) o(10) 你(20) 好(20) w(10) o(10) r(10) l(10) d(10)
+    expect(spans.length).toBe(12);
+
+    // Verify contiguity: no gaps between any adjacent spans
+    for (let i = 1; i < spans.length; i++) {
+      const prevEnd = spans[i - 1].left + spans[i - 1].width;
+      expect(spans[i].left).toBe(prevEnd);
+    }
+  });
+
+  it('addChar with Chinese characters accumulates correctly in addon', () => {
+    const { addon } = tracked();
+    for (const ch of '你好世界') {
+      addon.addChar(ch);
+    }
+    expect(addon.pendingText).toBe('你好世界');
+    expect(addon.hasPending).toBe(true);
+  });
+});
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 2: CJK text renders with correct double-width spacing
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 2: CJK double-width spacing', () => {
+  it('charCellWidth returns 2 for CJK Unified Ideographs (0x4E00-0x9FFF)', () => {
+    // Common Chinese characters
+    const chars = '中文测试你好世界天地人';
+    for (const ch of chars) {
+      expect(charCellWidth(null, ch)).toBe(2);
+    }
+  });
+
+  it('charCellWidth returns 2 for CJK Extension A (0x3400-0x4DBF)', () => {
+    expect(charCellWidth(null, '\u3400')).toBe(2); // First Extension A char
+    expect(charCellWidth(null, '\u4DB5')).toBe(2); // One of the last Extension A chars
+  });
+
+  it('charCellWidth returns 2 for CJK Radicals (0x2E80-0x2EFF)', () => {
+    expect(charCellWidth(null, '\u2E80')).toBe(2); // CJK Radical Repeat
+  });
+
+  it('charCellWidth returns 2 for fullwidth ASCII forms (0xFF01-0xFF5E)', () => {
+    expect(charCellWidth(null, '\uFF01')).toBe(2); // ！
+    expect(charCellWidth(null, '\uFF21')).toBe(2); // Ａ
+    expect(charCellWidth(null, '\uFF41')).toBe(2); // ａ
+    expect(charCellWidth(null, '\uFF10')).toBe(2); // ０
+  });
+
+  it('charCellWidth returns 2 for CJK Compatibility Ideographs (0xF900-0xFAFF)', () => {
+    expect(charCellWidth(null, '\uF900')).toBe(2);
+  });
+
+  it('stringCellWidth calculates correct visual width for CJK strings', () => {
+    expect(stringCellWidth(null, '你好')).toBe(4); // 2 + 2
+    expect(stringCellWidth(null, '你好世界')).toBe(8); // 4 * 2
+    expect(stringCellWidth(null, 'hi你好')).toBe(6); // 1 + 1 + 2 + 2
+    expect(stringCellWidth(null, '你a好b')).toBe(6); // 2 + 1 + 2 + 1
+    expect(stringCellWidth(null, 'abc你好def')).toBe(10); // 3 + 4 + 3
+  });
+
+  it('CJK span widths are exactly 2 * cellW pixels', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['你'], cellW: 8.4 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.width).toBe('16.8px'); // 2 * 8.4
+  });
+
+  it('ASCII span widths remain 1 * cellW pixels', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['a'], cellW: 8.4 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.width).toBe('8.4px'); // 1 * 8.4
+  });
+
+  it('terminal unicode addon is preferred over fallback when available', () => {
+    const mockTerminal = {
+      unicode: {
+        getStringCellWidth: (s: string) => {
+          // Custom width: treat 'W' as wide
+          return s === 'W' ? 2 : 1;
+        },
+      },
+    } as any;
+    expect(charCellWidth(mockTerminal, 'W')).toBe(2);
+    expect(charCellWidth(mockTerminal, 'a')).toBe(1);
+    // Fallback ignores terminal addon result for actual CJK
+    expect(charCellWidth(null, '你')).toBe(2);
+  });
+});
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 3: Japanese (こんにちは) and Korean (안녕하세요) input
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 3: Japanese and Korean input', () => {
+  describe('Japanese', () => {
+    it('charCellWidth returns 2 for Hiragana (0x3040-0x309F)', () => {
+      const hiragana = 'あいうえおかきくけこさしすせそ';
+      for (const ch of hiragana) {
+        expect(charCellWidth(null, ch)).toBe(2);
+      }
+    });
+
+    it('charCellWidth returns 2 for Katakana (0x30A0-0x30FF)', () => {
+      const katakana = 'アイウエオカキクケコサシスセソ';
+      for (const ch of katakana) {
+        expect(charCellWidth(null, ch)).toBe(2);
+      }
+    });
+
+    it('Japanese greeting renders with correct span positions', () => {
+      const container = document.createElement('div');
+      const text = 'こんにちは';
+      renderOverlay(container, makeParams({ lines: [text], cellW: 10 }));
+
+      const lineDiv = container.children[0] as HTMLDivElement;
+      const spans = getSpans(lineDiv);
+
+      expect(spans.length).toBe(5);
+      // こ(0,20) ん(20,20) に(40,20) ち(60,20) は(80,20)
+      expect(spans[0]).toEqual({ text: 'こ', left: 0, width: 20 });
+      expect(spans[1]).toEqual({ text: 'ん', left: 20, width: 20 });
+      expect(spans[2]).toEqual({ text: 'に', left: 40, width: 20 });
+      expect(spans[3]).toEqual({ text: 'ち', left: 60, width: 20 });
+      expect(spans[4]).toEqual({ text: 'は', left: 80, width: 20 });
+    });
+
+    it('mixed Japanese + ASCII positions correctly', () => {
+      const container = document.createElement('div');
+      const text = 'hello こんにちは';
+      renderOverlay(container, makeParams({ lines: [text], cellW: 10 }));
+
+      const lineDiv = container.children[0] as HTMLDivElement;
+      const spans = getSpans(lineDiv);
+
+      // h(0) e(10) l(20) l(30) o(40) space(50) こ(60) ん(80) に(100) ち(120) は(140)
+      expect(spans.length).toBe(11);
+      expect(spans[5]).toEqual({ text: ' ', left: 50, width: 10 }); // space
+      expect(spans[6]).toEqual({ text: 'こ', left: 60, width: 20 }); // first CJK after ASCII
+    });
+
+    it('addon handles Japanese input via addChar', () => {
+      const { addon } = tracked();
+      for (const ch of 'こんにちは') {
+        addon.addChar(ch);
+      }
+      expect(addon.pendingText).toBe('こんにちは');
+    });
+
+    it('addon handles Japanese input via appendText (paste)', () => {
+      const { addon } = tracked();
+      addon.appendText('こんにちは世界');
+      expect(addon.pendingText).toBe('こんにちは世界');
+      expect(addon.hasPending).toBe(true);
+    });
+
+    it('stringCellWidth correct for Japanese greeting', () => {
+      expect(stringCellWidth(null, 'こんにちは')).toBe(10); // 5 * 2
+    });
+  });
+
+  describe('Korean', () => {
+    it('charCellWidth returns 2 for Hangul Syllables (0xAC00-0xD7A3)', () => {
+      const hangul = '가나다라마바사아자차카타파하';
+      for (const ch of hangul) {
+        expect(charCellWidth(null, ch)).toBe(2);
+      }
+    });
+
+    it('charCellWidth returns 2 for Hangul Jamo (0x1100-0x115F)', () => {
+      expect(charCellWidth(null, '\u1100')).toBe(2); // ᄀ
+      expect(charCellWidth(null, '\u1112')).toBe(2); // ᄒ
+    });
+
+    it('Korean greeting renders with correct span positions', () => {
+      const container = document.createElement('div');
+      const text = '안녕하세요';
+      renderOverlay(container, makeParams({ lines: [text], cellW: 10 }));
+
+      const lineDiv = container.children[0] as HTMLDivElement;
+      const spans = getSpans(lineDiv);
+
+      expect(spans.length).toBe(5);
+      expect(spans[0]).toEqual({ text: '안', left: 0, width: 20 });
+      expect(spans[1]).toEqual({ text: '녕', left: 20, width: 20 });
+      expect(spans[2]).toEqual({ text: '하', left: 40, width: 20 });
+      expect(spans[3]).toEqual({ text: '세', left: 60, width: 20 });
+      expect(spans[4]).toEqual({ text: '요', left: 80, width: 20 });
+    });
+
+    it('addon handles Korean input', () => {
+      const { addon } = tracked();
+      addon.appendText('안녕하세요');
+      expect(addon.pendingText).toBe('안녕하세요');
+      expect(addon.hasPending).toBe(true);
+    });
+
+    it('removeChar removes Korean characters one at a time', () => {
+      const { addon } = tracked();
+      addon.appendText('안녕');
+      addon.removeChar();
+      expect(addon.pendingText).toBe('안');
+      addon.removeChar();
+      expect(addon.pendingText).toBe('');
+    });
+
+    it('stringCellWidth correct for Korean greeting', () => {
+      expect(stringCellWidth(null, '안녕하세요')).toBe(10); // 5 * 2
+    });
+  });
+
+  describe('mixed CJK scripts', () => {
+    it('Chinese + Japanese + Korean in one string', () => {
+      const text = '你好こんにちは안녕';
+      expect(stringCellWidth(null, text)).toBe(18); // 9 chars * 2 each
+
+      const container = document.createElement('div');
+      renderOverlay(container, makeParams({ lines: [text], cellW: 10 }));
+      const lineDiv = container.children[0] as HTMLDivElement;
+      const spans = getSpans(lineDiv);
+
+      // All 9 CJK chars: contiguous double-width spans
+      expect(spans.length).toBe(9);
+      let expectedLeft = 0;
+      for (const span of spans) {
+        expect(span.left).toBe(expectedLeft);
+        expect(span.width).toBe(20);
+        expectedLeft += 20;
+      }
+    });
+  });
+});
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 4: Cursor positions correctly after CJK characters
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 4: Cursor positioning after CJK', () => {
+  it('cursor after Chinese text on first line', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世界'],
+        startCol: 2,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // cursorCol = startCol(2) + stringCellWidth('你好世界')(8) = 10
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('100px'); // 10 * 10px
+  });
+
+  it('cursor after Japanese text on first line', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['こんにちは'],
+        startCol: 2,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // cursorCol = 2 + 10 = 12
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('120px');
+  });
+
+  it('cursor after Korean text on first line', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['안녕하세요'],
+        startCol: 2,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // cursorCol = 2 + 10 = 12
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('120px');
+  });
+
+  it('cursor after mixed ASCII + CJK', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['hi你好'],
+        startCol: 3,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // cursorCol = 3 + stringCellWidth('hi你好')(6) = 9
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('90px');
+  });
+
+  it('cursor on wrapped line after CJK text', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世界', '再见'],
+        startCol: 2,
+        cellW: 10,
+        cellH: 20,
+        showCursor: true,
+      })
+    );
+    // Last line is '再见', starts at col 0 (wrapped line), width = 4
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('40px'); // 0 + 4 = 4, * 10 = 40
+    expect(cursor.style.top).toBe('20px'); // row 1 * cellH
+  });
+
+  it('cursor hidden when it would exceed totalCols', () => {
+    const container = document.createElement('div');
+    // 4 CJK chars = 8 visual cols, startCol=73, cursorCol = 73 + 8 = 81 > 80
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世界'],
+        startCol: 73,
+        totalCols: 80,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // Should be 1 line div, no cursor span (cursor at col 81 >= totalCols 80)
+    // Actually cursor check is cursorCol < totalCols, so at 81 it's hidden
+    const children = container.children;
+    // If cursor is rendered, last child would be a cursor span
+    // With cursorCol 81 >= 80, cursor should NOT be rendered
+    expect(children.length).toBe(1); // only line div
+  });
+});
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 5: Long CJK input wraps at correct column boundary
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 5: CJK line wrapping at column boundaries', () => {
+  it('CJK chars wrap when they would overflow first line', () => {
+    const container = document.createElement('div');
+    // totalCols=10, startCol=2 → firstLineCols=8
+    // Each CJK char = 2 cols → first line fits 4 chars (8 cols)
+    // 5th char wraps to second line
+    const text = '你好世界啊'; // 5 chars = 10 visual cols
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世界', '啊'],
+        startCol: 2,
+        totalCols: 10,
+        cellW: 10,
+        cellH: 20,
+      })
+    );
+
+    expect(container.children.length).toBe(3); // 2 line divs + cursor
+
+    const line1 = container.children[0] as HTMLDivElement;
+    expect(line1.children.length).toBe(4); // 你好世界
+    expect(line1.style.left).toBe('20px'); // startCol * cellW
+
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line2.children.length).toBe(1); // 啊
+    expect(line2.style.left).toBe('0px'); // wrapped line starts at col 0
+    expect(line2.style.top).toBe('20px'); // second row
+  });
+
+  it('CJK char that would partially overflow stays on next line', () => {
+    // totalCols=9, startCol=2 → firstLineCols=7
+    // CJK chars are 2-wide. 3 chars = 6 cols (fits). 4th char = 8 cols > 7. Wraps.
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世', '界'],
+        startCol: 2,
+        totalCols: 9,
+        cellW: 10,
+      })
+    );
+
+    const line1 = container.children[0] as HTMLDivElement;
+    expect(line1.children.length).toBe(3); // 3 CJK chars fit (6 cols <= 7)
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line2.children.length).toBe(1); // 界 wraps
+  });
+
+  it('addon _render splits CJK text into visual lines correctly', () => {
+    // Narrow terminal: 10 cols, startCol=2 → firstLineCols=8 → 4 CJK chars
+    const mock = createMockTerminal({
+      buffer: { lines: ['$ '] },
+      cols: 10,
+    });
+    const addon = new ZerolagInputAddon({
+      prompt: { type: 'character', char: '$', offset: 2 },
+    });
+    mock.terminal.loadAddon(addon);
+    cleanups.push(() => {
+      addon.dispose();
+      mock.cleanup();
+    });
+
+    // Type 6 CJK chars (12 visual cols)
+    for (const ch of '你好世界再见') {
+      addon.addChar(ch);
+    }
+    expect(addon.pendingText).toBe('你好世界再见');
+    expect(addon.hasPending).toBe(true);
+  });
+
+  it('mixed ASCII + CJK wraps correctly', () => {
+    const container = document.createElement('div');
+    // totalCols=10, startCol=2 → firstLineCols=8
+    // 'ab' = 2 cols, '你好' = 4 cols, 'cd' = 2 cols → total 8 cols (fits line 1)
+    // '世' = 2 cols → wraps to line 2
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['ab你好cd', '世'],
+        startCol: 2,
+        totalCols: 10,
+        cellW: 10,
+      })
+    );
+
+    const line1 = container.children[0] as HTMLDivElement;
+    expect(line1.children.length).toBe(6); // a,b,你,好,c,d
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line2.children.length).toBe(1); // 世
+  });
+
+  it('odd column count: CJK char does not split across boundary', () => {
+    // totalCols=11, startCol=2 → firstLineCols=9
+    // 4 CJK chars = 8 cols (fits). 5th CJK = 10 cols > 9 (wraps).
+    // 1 empty column remains on first line (CJK can't fit in 1 col).
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世界', '啊'],
+        startCol: 2,
+        totalCols: 11,
+        cellW: 10,
+      })
+    );
+
+    const line1 = container.children[0] as HTMLDivElement;
+    expect(line1.children.length).toBe(4); // 4 chars = 8 cols, 5th would need 10 > 9
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line2.children.length).toBe(1);
+  });
+
+  it('CJK fills entire wrapped line', () => {
+    const container = document.createElement('div');
+    // totalCols=6, startCol=0 → firstLineCols=6
+    // 3 CJK chars = 6 cols (fills line 1), next 3 fill line 2
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好世', '界再见'],
+        startCol: 0,
+        totalCols: 6,
+        cellW: 10,
+      })
+    );
+
+    const line1 = container.children[0] as HTMLDivElement;
+    expect(line1.children.length).toBe(3);
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line2.children.length).toBe(3);
+  });
+});
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 6: Existing ASCII input is unaffected
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 6: ASCII input — no regression', () => {
+  it('ASCII characters still get 1-cell-wide spans', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['abcdef'], cellW: 10 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const spans = getSpans(lineDiv);
+
+    for (const span of spans) {
+      expect(span.width).toBe(10); // 1 * cellW
+    }
+  });
+
+  it('ASCII span positions are sequential at 1-cell intervals', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['xyz'], cellW: 8.4 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const spans = getSpans(lineDiv);
+
+    expect(spans[0].left).toBe(0);
+    expect(spans[1].left).toBeCloseTo(8.4, 5);
+    expect(spans[2].left).toBeCloseTo(16.8, 5);
+  });
+
+  it('ASCII cursor positions at correct column', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['hello'],
+        startCol: 3,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // cursor at startCol(3) + 5 = 8
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('80px');
+  });
+
+  it('charCellWidth returns 1 for all printable ASCII', () => {
+    for (let code = 32; code < 127; code++) {
+      const ch = String.fromCharCode(code);
+      expect(charCellWidth(null, ch)).toBe(1);
+    }
+  });
+
+  it('stringCellWidth equals length for pure ASCII', () => {
+    expect(stringCellWidth(null, 'hello world')).toBe(11);
+    expect(stringCellWidth(null, 'test123!@#')).toBe(10);
+    expect(stringCellWidth(null, '')).toBe(0);
+  });
+
+  it('ASCII multi-line wrapping is unaffected', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['abcde', 'fgh'],
+        startCol: 5,
+        totalCols: 10,
+        cellW: 10,
+        cellH: 20,
+      })
+    );
+
+    expect(container.children.length).toBe(3); // 2 lines + cursor
+    const line1 = container.children[0] as HTMLDivElement;
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line1.children.length).toBe(5);
+    expect(line2.children.length).toBe(3);
+    expect(line1.style.left).toBe('50px'); // startCol * cellW
+    expect(line2.style.left).toBe('0px');
+  });
+
+  it('addon addChar/removeChar/clear work for ASCII', () => {
+    const { addon } = tracked();
+    addon.addChar('h');
+    addon.addChar('e');
+    addon.addChar('l');
+    addon.addChar('l');
+    addon.addChar('o');
+    expect(addon.pendingText).toBe('hello');
+
+    addon.removeChar();
+    expect(addon.pendingText).toBe('hell');
+
+    addon.clear();
+    expect(addon.pendingText).toBe('');
+    expect(addon.hasPending).toBe(false);
+  });
+});
+
+// ═══════════════════════════════════════════════════════════════════════
+// TEST 7: Teammate terminal panels render CJK correctly
+//
+// Teammate panels share the global xterm-zerolag-input.js vendor bundle
+// (loaded via <script> in index.html). This is the IIFE build of the
+// same package source tested in tests 1-6. The build pipeline is:
+//   src/overlay-renderer.ts → tsup → dist/index.global.js → vendor copy
+//
+// Since all terminals (main + teammate) use the same LocalEchoOverlay
+// class from the global scope, CJK correctness is guaranteed by:
+// (a) The package source handles CJK correctly (tests 1-6 above)
+// (b) The IIFE build bundles the exact same charCellWidth / makeLine code
+//
+// These tests verify the exported module includes CJK-aware functions
+// and that teammate-style terminal instances work identically.
+// ═══════════════════════════════════════════════════════════════════════
+
+describe('Test 7: Teammate terminal panels render CJK correctly', () => {
+  it('package exports charCellWidth and stringCellWidth', () => {
+    // These are the CJK-aware functions that the IIFE build exposes
+    expect(typeof charCellWidth).toBe('function');
+    expect(typeof stringCellWidth).toBe('function');
+  });
+
+  it('ZerolagInputAddon (used by LocalEchoOverlay) handles CJK in teammate terminals', () => {
+    // Simulate a teammate terminal panel: separate terminal instance,
+    // same addon class, different prompt character
+    const mock = createMockTerminal({
+      buffer: { lines: ['❯ '] },
+      cols: 40,
+    });
+    const addon = new ZerolagInputAddon({
+      prompt: { type: 'character', char: '❯', offset: 2 },
+    });
+    mock.terminal.loadAddon(addon);
+    cleanups.push(() => {
+      addon.dispose();
+      mock.cleanup();
+    });
+
+    // Type CJK in teammate terminal
+    for (const ch of '你好世界') {
+      addon.addChar(ch);
+    }
+    expect(addon.pendingText).toBe('你好世界');
+    expect(addon.hasPending).toBe(true);
+  });
+
+  it('teammate terminal with narrow width wraps CJK correctly', () => {
+    // Teammate panels are often narrower (sidebar, split view)
+    const mock = createMockTerminal({
+      buffer: { lines: ['❯ '] },
+      cols: 12, // narrow panel
+    });
+    const addon = new ZerolagInputAddon({
+      prompt: { type: 'character', char: '❯', offset: 2 },
+    });
+    mock.terminal.loadAddon(addon);
+    cleanups.push(() => {
+      addon.dispose();
+      mock.cleanup();
+    });
+
+    // Type 6 CJK chars (12 visual cols) with startCol=2 → only 10 available
+    // First line: 5 CJK = 10 cols. 6th wraps.
+    for (const ch of '你好世界再见') {
+      addon.addChar(ch);
+    }
+    expect(addon.pendingText).toBe('你好世界再见');
+  });
+
+  it('mixed CJK scripts render identically in teammate and main terminals', () => {
+    // Verify that the same input produces the same overlay output
+    // regardless of which terminal instance it's on
+    const text = '你好こんにちは안녕';
+
+    // "Main" terminal
+    const container1 = document.createElement('div');
+    renderOverlay(container1, makeParams({ lines: [text], cellW: 10 }));
+    const line1 = container1.children[0] as HTMLDivElement;
+    const spans1 = getSpans(line1);
+
+    // "Teammate" terminal (same params — same bundle)
+    const container2 = document.createElement('div');
+    renderOverlay(container2, makeParams({ lines: [text], cellW: 10 }));
+    const line2 = container2.children[0] as HTMLDivElement;
+    const spans2 = getSpans(line2);
+
+    // Identical rendering
+    expect(spans1.length).toBe(spans2.length);
+    for (let i = 0; i < spans1.length; i++) {
+      expect(spans1[i]).toEqual(spans2[i]);
+    }
+  });
+
+  it('CJK rendering correct in overlay with teammate-typical prompt (❯)', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['こんにちは世界'],
+        startCol: 2, // after ❯ prompt
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const spans = getSpans(lineDiv);
+    expect(spans.length).toBe(7);
+
+    // All CJK, all double-width, contiguous
+    let expectedLeft = 0;
+    for (const span of spans) {
+      expect(span.left).toBe(expectedLeft);
+      expect(span.width).toBe(20);
+      expectedLeft += 20;
+    }
+
+    // Cursor position: startCol(2) + 14 visual cols = 16
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('160px');
+  });
+});

packages/xterm-zerolag-input/test/helpers.ts

@@ -31,6 +31,10 @@ interface MockTerminalOptions {
     };
     cellWidth?: number;
     cellHeight?: number;
+    /** Device-pixel char top offset (for charTop calculation). Default: 0 */
+    deviceCharTop?: number;
+    /** Device-pixel char height (for charHeight calculation). Default: cellHeight * dpr */
+    deviceCharHeight?: number;
 }
 
 export function createMockTerminal(opts: MockTerminalOptions = {}) {
@@ -95,6 +99,12 @@ export function createMockTerminal(opts: MockTerminalOptions = {}) {
                     css: {
                         cell: { width: cellW, height: cellH },
                     },
+                    device: {
+                        char: {
+                            top: opts.deviceCharTop ?? 0,
+                            height: opts.deviceCharHeight ?? cellH,
+                        },
+                    },
                 },
             },
         },

packages/xterm-zerolag-input/test/overlay-renderer.test.ts

@@ -1,169 +1,420 @@
 import { describe, it, expect } from 'vitest';
-import { renderOverlay } from '../src/overlay-renderer.js';
+import { renderOverlay, charCellWidth, stringCellWidth } from '../src/overlay-renderer.js';
 import type { RenderParams, FontStyle } from '../src/types.js';
 
 const FONT: FontStyle = {
-    fontFamily: 'monospace',
-    fontSize: '14px',
-    fontWeight: 'normal',
-    color: '#eeeeee',
-    backgroundColor: '#0d0d0d',
-    letterSpacing: '',
+  fontFamily: 'monospace',
+  fontSize: '14px',
+  fontWeight: 'normal',
+  color: '#eeeeee',
+  backgroundColor: '#0d0d0d',
+  letterSpacing: '',
 };
 
 function makeParams(overrides: Partial<RenderParams> = {}): RenderParams {
-    return {
-        lines: ['hello'],
-        startCol: 2,
-        totalCols: 80,
-        cellW: 8.4,
-        cellH: 17,
-        promptRow: 10,
-        font: FONT,
-        showCursor: true,
-        cursorColor: '#e0e0e0',
-        ...overrides,
-    };
+  return {
+    lines: ['hello'],
+    startCol: 2,
+    totalCols: 80,
+    cellW: 8.4,
+    cellH: 17,
+    charTop: 2,
+    charHeight: 14,
+    promptRow: 10,
+    font: FONT,
+    showCursor: true,
+    cursorColor: '#e0e0e0',
+    ...overrides,
+  };
 }
 
 describe('renderOverlay', () => {
-    it('positions container at prompt row', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ promptRow: 5 }));
-        expect(container.style.top).toBe((5 * 17) + 'px');
-        expect(container.style.left).toBe('0px');
-    });
+  it('positions container at prompt row', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ promptRow: 5 }));
+    expect(container.style.top).toBe(5 * 17 + 'px');
+    expect(container.style.left).toBe('0px');
+  });
 
-    it('creates per-character spans in a line div', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ lines: ['abc'] }));
+  it('creates per-character spans in a line div', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['abc'] }));
 
-        // Line div + cursor span
-        expect(container.children.length).toBe(2);
+    // Line div + cursor span
+    expect(container.children.length).toBe(2);
 
-        const lineDiv = container.children[0] as HTMLDivElement;
-        expect(lineDiv.children.length).toBe(3); // a, b, c
+    const lineDiv = container.children[0] as HTMLDivElement;
+    expect(lineDiv.children.length).toBe(3); // a, b, c
 
-        const spanA = lineDiv.children[0] as HTMLSpanElement;
-        expect(spanA.textContent).toBe('a');
-        expect(spanA.style.left).toBe('0px');
+    const spanA = lineDiv.children[0] as HTMLSpanElement;
+    expect(spanA.textContent).toBe('a');
+    expect(spanA.style.left).toBe('0px');
 
-        const spanB = lineDiv.children[1] as HTMLSpanElement;
-        expect(spanB.textContent).toBe('b');
-        expect(spanB.style.left).toBe('8.4px');
+    const spanB = lineDiv.children[1] as HTMLSpanElement;
+    expect(spanB.textContent).toBe('b');
+    expect(spanB.style.left).toBe('8.4px');
 
-        const spanC = lineDiv.children[2] as HTMLSpanElement;
-        expect(spanC.textContent).toBe('c');
-        expect(spanC.style.left).toBe('16.8px');
-    });
+    const spanC = lineDiv.children[2] as HTMLSpanElement;
+    expect(spanC.textContent).toBe('c');
+    expect(spanC.style.left).toBe('16.8px');
+  });
 
-    it('sets span width to cellW', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ lines: ['x'], cellW: 9.5 }));
-        const lineDiv = container.children[0] as HTMLDivElement;
-        const span = lineDiv.children[0] as HTMLSpanElement;
-        expect(span.style.width).toBe('9.5px');
-    });
+  it('sets span width to cellW', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['x'], cellW: 9.5 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.width).toBe('9.5px');
+  });
 
-    it('applies font styles to spans', () => {
-        const font: FontStyle = {
-            fontFamily: 'Fira Code',
-            fontSize: '16px',
-            fontWeight: 'bold',
-            color: '#ff0000',
-            backgroundColor: '#000000',
-            letterSpacing: '0.5px',
-        };
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ lines: ['A'], font }));
+  it('applies font styles to spans', () => {
+    const font: FontStyle = {
+      fontFamily: 'Fira Code',
+      fontSize: '16px',
+      fontWeight: 'bold',
+      color: '#ff0000',
+      backgroundColor: '#000000',
+      letterSpacing: '0.5px',
+    };
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['A'], font }));
 
-        const lineDiv = container.children[0] as HTMLDivElement;
-        // jsdom normalizes hex to rgb()
-        expect(lineDiv.style.backgroundColor).toBe('rgb(0, 0, 0)');
+    const lineDiv = container.children[0] as HTMLDivElement;
+    // jsdom normalizes hex to rgb()
+    expect(lineDiv.style.backgroundColor).toBe('rgb(0, 0, 0)');
 
-        const span = lineDiv.children[0] as HTMLSpanElement;
-        expect(span.style.fontFamily).toBe('Fira Code');
-        expect(span.style.fontSize).toBe('16px');
-        expect(span.style.fontWeight).toBe('bold');
-        expect(span.style.color).toBe('rgb(255, 0, 0)');
-        expect(span.style.letterSpacing).toBe('0.5px');
-    });
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.fontFamily).toBe('Fira Code');
+    expect(span.style.fontSize).toBe('16px');
+    expect(span.style.fontWeight).toBe('bold');
+    expect(span.style.color).toBe('rgb(255, 0, 0)');
+    expect(span.style.letterSpacing).toBe('0.5px');
+  });
 
-    it('offsets first line by startCol', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ lines: ['hi'], startCol: 5, cellW: 10 }));
-        const lineDiv = container.children[0] as HTMLDivElement;
-        // First line left = startCol * cellW
-        expect(lineDiv.style.left).toBe('50px');
-    });
+  it('offsets first line by startCol', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['hi'], startCol: 5, cellW: 10 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    // First line left = startCol * cellW
+    expect(lineDiv.style.left).toBe('50px');
+  });
 
-    it('renders cursor at end of text', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({
-            lines: ['ab'],
-            startCol: 3,
-            cellW: 10,
-            cellH: 20,
-            showCursor: true,
-            cursorColor: '#ff00ff',
-        }));
+  it('renders cursor at end of text', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['ab'],
+        startCol: 3,
+        cellW: 10,
+        cellH: 20,
+        showCursor: true,
+        cursorColor: '#ff00ff',
+      })
+    );
 
-        // Last child is cursor (after line div)
-        const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
-        // cursorCol = startCol(3) + text.length(2) = 5
-        expect(cursor.style.left).toBe('50px');
-        expect(cursor.style.width).toBe('10px');
-        expect(cursor.style.height).toBe('20px');
-        // jsdom normalizes hex to rgb()
-        expect(cursor.style.backgroundColor).toBe('rgb(255, 0, 255)');
-    });
+    // Last child is cursor (after line div)
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    // cursorCol = startCol(3) + text.length(2) = 5
+    expect(cursor.style.left).toBe('50px');
+    expect(cursor.style.width).toBe('10px');
+    expect(cursor.style.height).toBe('20px');
+    // jsdom normalizes hex to rgb()
+    expect(cursor.style.backgroundColor).toBe('rgb(255, 0, 255)');
+  });
 
-    it('does not render cursor when showCursor is false', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ lines: ['ab'], showCursor: false }));
-        // Only line div, no cursor
-        expect(container.children.length).toBe(1);
-    });
+  it('does not render cursor when showCursor is false', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['ab'], showCursor: false }));
+    // Only line div, no cursor
+    expect(container.children.length).toBe(1);
+  });
 
-    it('renders multi-line text', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({
-            lines: ['first', 'second'],
-            startCol: 5,
-            cellW: 10,
-            cellH: 20,
-        }));
+  it('renders multi-line text', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['first', 'second'],
+        startCol: 5,
+        cellW: 10,
+        cellH: 20,
+      })
+    );
 
-        // 2 line divs + cursor
-        expect(container.children.length).toBe(3);
+    // 2 line divs + cursor
+    expect(container.children.length).toBe(3);
 
-        const line1 = container.children[0] as HTMLDivElement;
-        expect(line1.style.left).toBe('50px'); // startCol * cellW
-        expect(line1.style.top).toBe('0px');
-        expect(line1.children.length).toBe(5); // 'first'
+    const line1 = container.children[0] as HTMLDivElement;
+    expect(line1.style.left).toBe('50px'); // startCol * cellW
+    expect(line1.style.top).toBe('0px');
+    expect(line1.children.length).toBe(5); // 'first'
 
-        const line2 = container.children[1] as HTMLDivElement;
-        expect(line2.style.left).toBe('0px'); // wrapped lines start at col 0
-        expect(line2.style.top).toBe('20px'); // second row
-        expect(line2.children.length).toBe(6); // 'second'
-    });
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line2.style.left).toBe('0px'); // wrapped lines start at col 0
+    expect(line2.style.top).toBe('20px'); // second row
+    expect(line2.children.length).toBe(6); // 'second'
+  });
 
-    it('clears previous content on re-render', () => {
-        const container = document.createElement('div');
-        renderOverlay(container, makeParams({ lines: ['abc'] }));
-        expect(container.children.length).toBe(2); // line + cursor
+  it('clears previous content on re-render', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['abc'] }));
+    expect(container.children.length).toBe(2); // line + cursor
 
-        renderOverlay(container, makeParams({ lines: ['xy'] }));
-        expect(container.children.length).toBe(2); // line + cursor (rebuilt)
+    renderOverlay(container, makeParams({ lines: ['xy'] }));
+    expect(container.children.length).toBe(2); // line + cursor (rebuilt)
 
-        const lineDiv = container.children[0] as HTMLDivElement;
-        expect(lineDiv.children.length).toBe(2); // x, y
-    });
+    const lineDiv = container.children[0] as HTMLDivElement;
+    expect(lineDiv.children.length).toBe(2); // x, y
+  });
 
-    it('shows container (display not none)', () => {
-        const container = document.createElement('div');
-        container.style.display = 'none';
-        renderOverlay(container, makeParams());
-        expect(container.style.display).toBe('');
-    });
+  it('shows container (display not none)', () => {
+    const container = document.createElement('div');
+    container.style.display = 'none';
+    renderOverlay(container, makeParams());
+    expect(container.style.display).toBe('');
+  });
+
+  // ─── Anti-flicker / compositing seam tests ────────────────────
+
+  it('line div height extends 1px past cellH to cover compositing seam', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['abc'], cellH: 19 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    // cellH + 1 = 20px — the extra 1px covers the compositing seam
+    expect(lineDiv.style.height).toBe('20px');
+  });
+
+  it('line div height is cellH+1 for various cell heights', () => {
+    for (const cellH of [15, 17, 19, 22]) {
+      const container = document.createElement('div');
+      renderOverlay(container, makeParams({ lines: ['x'], cellH }));
+      const lineDiv = container.children[0] as HTMLDivElement;
+      expect(lineDiv.style.height).toBe(cellH + 1 + 'px');
+    }
+  });
+
+  it('multi-line overlay has cellH+1 height on each line div', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['first', 'second'],
+        cellH: 19,
+      })
+    );
+    const line1 = container.children[0] as HTMLDivElement;
+    const line2 = container.children[1] as HTMLDivElement;
+    expect(line1.style.height).toBe('20px');
+    expect(line2.style.height).toBe('20px');
+  });
+
+  // ─── Span vertical centering tests ────────────────────────────
+
+  it('span uses full cellH for height and lineHeight (CSS centering)', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['a'], cellH: 19 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.height).toBe('19px');
+    expect(span.style.lineHeight).toBe('19px');
+  });
+
+  it('span top is 0px (no vertical offset / no transform)', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['a'], cellH: 19 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.top).toBe('0px');
+    // No translateY transform — sub-pixel overhang causes artifacts
+    expect(span.style.transform).toBe('');
+  });
+
+  // ─── Font rendering tests ─────────────────────────────────────
+
+  it('span disables ligatures via font-feature-settings', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['fi'] }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    // Check cssText includes the ligature-disabling settings
+    // jsdom may normalize whitespace; check that both liga and calt are disabled
+    expect(span.style.cssText).toContain('font-feature-settings:');
+    expect(span.style.cssText).toContain("'liga' 0");
+    expect(span.style.cssText).toContain("'calt' 0");
+  });
+
+  it('span has text-align: center for glyph centering', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['m'] }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.textAlign).toBe('center');
+  });
+
+  it('span has pointer-events: none', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['a'] }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    const span = lineDiv.children[0] as HTMLSpanElement;
+    expect(span.style.pointerEvents).toBe('none');
+  });
+
+  // ─── Multi-line cursor positioning ────────────────────────────
+
+  it('cursor on wrapped line uses col 0 as base', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['first', 'ab'],
+        startCol: 5,
+        cellW: 10,
+        cellH: 20,
+        showCursor: true,
+      })
+    );
+    // Cursor at end of second line: col = 0 + 2 = 2
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('20px'); // 2 * 10
+    expect(cursor.style.top).toBe('20px'); // row 1 * cellH
+  });
+
+  // ─── charTop/charHeight passed through ────────────────────────
+
+  it('accepts charTop and charHeight params without error', () => {
+    const container = document.createElement('div');
+    expect(() =>
+      renderOverlay(
+        container,
+        makeParams({
+          lines: ['test'],
+          charTop: 2,
+          charHeight: 14,
+        })
+      )
+    ).not.toThrow();
+    expect(container.children.length).toBeGreaterThan(0);
+  });
+
+  // ─── Line div positioning regression ──────────────────────────
+
+  it('line div background color matches font.backgroundColor', () => {
+    const container = document.createElement('div');
+    const font: FontStyle = { ...FONT, backgroundColor: '#1a1a1a' };
+    renderOverlay(container, makeParams({ lines: ['x'], font }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    // jsdom normalizes hex to rgb()
+    expect(lineDiv.style.backgroundColor).toBe('rgb(26, 26, 26)');
+  });
+
+  it('empty line produces line div with no spans', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: [''] }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    expect(lineDiv.children.length).toBe(0);
+  });
+
+  // ─── CJK wide character support ───────────────────────────────
+
+  it('CJK characters get double-width spans', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['a你b'], cellW: 10 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    expect(lineDiv.children.length).toBe(3);
+
+    const spanA = lineDiv.children[0] as HTMLSpanElement;
+    expect(spanA.textContent).toBe('a');
+    expect(spanA.style.left).toBe('0px');
+    expect(spanA.style.width).toBe('10px'); // 1 cell
+
+    const spanCJK = lineDiv.children[1] as HTMLSpanElement;
+    expect(spanCJK.textContent).toBe('你');
+    expect(spanCJK.style.left).toBe('10px'); // col 1
+    expect(spanCJK.style.width).toBe('20px'); // 2 cells
+
+    const spanB = lineDiv.children[2] as HTMLSpanElement;
+    expect(spanB.textContent).toBe('b');
+    expect(spanB.style.left).toBe('30px'); // col 3
+    expect(spanB.style.width).toBe('10px'); // 1 cell
+  });
+
+  it('cursor position accounts for CJK width', () => {
+    const container = document.createElement('div');
+    renderOverlay(
+      container,
+      makeParams({
+        lines: ['你好'],
+        startCol: 2,
+        cellW: 10,
+        showCursor: true,
+      })
+    );
+    // 你(2) + 好(2) = 4 visual cols, cursor at startCol(2) + 4 = 6
+    const cursor = container.children[container.children.length - 1] as HTMLSpanElement;
+    expect(cursor.style.left).toBe('60px');
+  });
+
+  it('mixed ASCII and CJK characters position correctly', () => {
+    const container = document.createElement('div');
+    renderOverlay(container, makeParams({ lines: ['hi你'], cellW: 8 }));
+    const lineDiv = container.children[0] as HTMLDivElement;
+    // h(col 0), i(col 1), 你(col 2, width 2)
+    const spanH = lineDiv.children[0] as HTMLSpanElement;
+    expect(spanH.style.left).toBe('0px');
+    const spanI = lineDiv.children[1] as HTMLSpanElement;
+    expect(spanI.style.left).toBe('8px');
+    const spanCJK = lineDiv.children[2] as HTMLSpanElement;
+    expect(spanCJK.style.left).toBe('16px');
+    expect(spanCJK.style.width).toBe('16px');
+  });
+});
+
+describe('charCellWidth', () => {
+  it('returns 1 for ASCII characters', () => {
+    expect(charCellWidth(null, 'a')).toBe(1);
+    expect(charCellWidth(null, '!')).toBe(1);
+    expect(charCellWidth(null, ' ')).toBe(1);
+  });
+
+  it('returns 2 for CJK ideographs', () => {
+    expect(charCellWidth(null, '你')).toBe(2);
+    expect(charCellWidth(null, '好')).toBe(2);
+    expect(charCellWidth(null, '中')).toBe(2);
+  });
+
+  it('returns 2 for Japanese hiragana', () => {
+    expect(charCellWidth(null, 'こ')).toBe(2);
+    expect(charCellWidth(null, 'ん')).toBe(2);
+  });
+
+  it('returns 2 for Korean syllables', () => {
+    expect(charCellWidth(null, '안')).toBe(2);
+    expect(charCellWidth(null, '녕')).toBe(2);
+  });
+
+  it('returns 2 for fullwidth forms', () => {
+    expect(charCellWidth(null, '\uff01')).toBe(2); // ！
+    expect(charCellWidth(null, '\uff21')).toBe(2); // Ａ
+  });
+
+  it('uses terminal unicode addon when available', () => {
+    const mockTerminal = {
+      unicode: { getStringCellWidth: (s: string) => (s === 'W' ? 2 : 1) },
+    } as any;
+    expect(charCellWidth(mockTerminal, 'W')).toBe(2);
+    expect(charCellWidth(mockTerminal, 'n')).toBe(1);
+  });
+});
+
+describe('stringCellWidth', () => {
+  it('sums individual character widths', () => {
+    expect(stringCellWidth(null, 'abc')).toBe(3);
+    expect(stringCellWidth(null, '你好')).toBe(4);
+    expect(stringCellWidth(null, 'a你b')).toBe(4);
+  });
+
+  it('returns 0 for empty string', () => {
+    expect(stringCellWidth(null, '')).toBe(0);
+  });
 });

packages/xterm-zerolag-input/tsup.config.ts

@@ -0,0 +1,33 @@
+import { defineConfig } from 'tsup';
+
+export default defineConfig([
+  // Standard builds (CJS + ESM + DTS)
+  {
+    entry: ['src/index.ts'],
+    format: ['cjs', 'esm'],
+    dts: true,
+    clean: true,
+  },
+  // IIFE build for browser <script> tag usage
+  {
+    entry: ['src/index.ts'],
+    format: ['iife'],
+    globalName: 'XtermZerolagInput',
+    outDir: 'dist',
+    // Append global aliases so app.js can access classes directly
+    footer: {
+      js: [
+        '// Global aliases for browser usage',
+        'if(typeof window!=="undefined"){',
+        '  window.ZerolagInputAddon=XtermZerolagInput.ZerolagInputAddon;',
+        '  window.LocalEchoOverlay=class extends XtermZerolagInput.ZerolagInputAddon{',
+        '    constructor(terminal){',
+        '      super({prompt:{type:"character",char:"\\u276f",offset:2}});',
+        '      this.activate(terminal);',
+        '    }',
+        '  };',
+        '}',
+      ].join('\n'),
+    },
+  },
+]);

scripts/build.mjs

@@ -8,12 +8,15 @@
  *   2. Copy static assets (web/public, templates)
  *   3. Build vendor xterm bundles
  *   4. Minify frontend assets (app.js, styles.css, mobile.css)
- *   5. Compress with gzip + brotli
+ *   5. Content-hash cache busting (rename assets, rewrite index.html)
+ *   6. Compress with gzip + brotli
  */
 
 import { execSync } from 'child_process';
+import { appendFileSync, readFileSync, writeFileSync, renameSync } from 'fs';
+import { createHash } from 'crypto';
 import { fileURLToPath } from 'url';
-import { join } from 'path';
+import { join, extname, basename, dirname } from 'path';
 
 const ROOT = join(fileURLToPath(import.meta.url), '..', '..');
 
@@ -26,24 +29,97 @@ function run(label, cmd) {
 run('tsc', 'tsc');
 run('chmod dist/index.js', 'chmod +x dist/index.js');
 
-// 2. Copy static assets
+// 2. Copy static assets (clean first to remove stale hashed files from previous builds)
+run('clean public', 'rm -rf dist/web/public');
 run('prepare dirs', 'mkdir -p dist/web dist/templates dist/web/public/vendor');
 run('copy web assets', 'cp -r src/web/public dist/web/');
 run('copy template', 'cp src/templates/case-template.md dist/templates/');
 
-// 3. Vendor xterm bundles
-run('xterm css', 'cp node_modules/xterm/css/xterm.css dist/web/public/vendor/');
-run('xterm js', 'npx esbuild node_modules/xterm/lib/xterm.js --minify --outfile=dist/web/public/vendor/xterm.min.js');
-run('xterm-addon-fit', 'npx esbuild node_modules/xterm-addon-fit/lib/xterm-addon-fit.js --minify --outfile=dist/web/public/vendor/xterm-addon-fit.min.js');
-run('xterm-addon-webgl', 'cp node_modules/xterm-addon-webgl/lib/xterm-addon-webgl.js dist/web/public/vendor/xterm-addon-webgl.min.js');
-run('xterm-addon-unicode11', 'npx esbuild node_modules/xterm-addon-unicode11/lib/xterm-addon-unicode11.js --minify --outfile=dist/web/public/vendor/xterm-addon-unicode11.min.js');
+// 3. Vendor xterm bundles (xterm.js 6.x — @xterm scoped packages)
+run('xterm css', 'cp node_modules/@xterm/xterm/css/xterm.css dist/web/public/vendor/');
+run('xterm js', 'npx esbuild node_modules/@xterm/xterm/lib/xterm.js --minify --outfile=dist/web/public/vendor/xterm.min.js');
+run('xterm-addon-fit', 'npx esbuild node_modules/@xterm/addon-fit/lib/addon-fit.js --minify --outfile=dist/web/public/vendor/xterm-addon-fit.min.js');
+run('xterm-addon-webgl', 'cp node_modules/@xterm/addon-webgl/lib/addon-webgl.js dist/web/public/vendor/xterm-addon-webgl.min.js');
+run('xterm-addon-unicode11', 'npx esbuild node_modules/@xterm/addon-unicode11/lib/addon-unicode11.js --minify --outfile=dist/web/public/vendor/xterm-addon-unicode11.min.js');
+run('xterm-zerolag-input', 'npx esbuild packages/xterm-zerolag-input/src/zerolag-input-addon.ts --bundle --minify --format=iife --global-name=XtermZerolagInput --outfile=dist/web/public/vendor/xterm-zerolag-input.js');
+
+// Append global aliases so app.js can use `new LocalEchoOverlay(terminal)`
+appendFileSync(
+  join(ROOT, 'dist/web/public/vendor/xterm-zerolag-input.js'),
+  '\n// Global aliases for browser usage\n' +
+  'if(typeof window!=="undefined"){' +
+    'window.ZerolagInputAddon=XtermZerolagInput.ZerolagInputAddon;' +
+    'window.LocalEchoOverlay=class extends XtermZerolagInput.ZerolagInputAddon{' +
+      'constructor(terminal){' +
+        'super({prompt:{type:"character",char:"\\u276f",offset:2}});' +
+        'this.activate(terminal);' +
+      '}' +
+    '};' +
+  '}\n'
+);
 
 // 4. Minify frontend assets
-run('minify app.js', 'npx esbuild dist/web/public/app.js --minify --drop:console --outfile=dist/web/public/app.js --allow-overwrite');
+run('minify input-cjk.js', 'npx esbuild dist/web/public/input-cjk.js --minify --outfile=dist/web/public/input-cjk.js --allow-overwrite');
+run('minify app.js', 'npx esbuild dist/web/public/app.js --minify --outfile=dist/web/public/app.js --allow-overwrite');
+run('minify terminal-ui.js', 'npx esbuild dist/web/public/terminal-ui.js --minify --outfile=dist/web/public/terminal-ui.js --allow-overwrite');
+run('minify respawn-ui.js', 'npx esbuild dist/web/public/respawn-ui.js --minify --outfile=dist/web/public/respawn-ui.js --allow-overwrite');
+run('minify ralph-panel.js', 'npx esbuild dist/web/public/ralph-panel.js --minify --outfile=dist/web/public/ralph-panel.js --allow-overwrite');
+run('minify settings-ui.js', 'npx esbuild dist/web/public/settings-ui.js --minify --outfile=dist/web/public/settings-ui.js --allow-overwrite');
+run('minify panels-ui.js', 'npx esbuild dist/web/public/panels-ui.js --minify --outfile=dist/web/public/panels-ui.js --allow-overwrite');
+run('minify session-ui.js', 'npx esbuild dist/web/public/session-ui.js --minify --outfile=dist/web/public/session-ui.js --allow-overwrite');
 run('minify styles.css', 'npx esbuild dist/web/public/styles.css --minify --outfile=dist/web/public/styles.css --allow-overwrite');
 run('minify mobile.css', 'npx esbuild dist/web/public/mobile.css --minify --outfile=dist/web/public/mobile.css --allow-overwrite');
 
-// 5. Compress with gzip + brotli
+// 5. Content-hash cache busting
+console.log('\n[build] content-hash cache busting');
+{
+  const distPublic = join(ROOT, 'dist/web/public');
+  const HASHABLE = [
+    'styles.css',
+    'mobile.css',
+    'constants.js',
+    'mobile-handlers.js',
+    'voice-input.js',
+    'notification-manager.js',
+    'keyboard-accessory.js',
+    'input-cjk.js',
+    'app.js',
+    'terminal-ui.js',
+    'respawn-ui.js',
+    'ralph-panel.js',
+    'settings-ui.js',
+    'panels-ui.js',
+    'session-ui.js',
+    'ralph-wizard.js',
+    'api-client.js',
+    'subagent-windows.js',
+    'vendor/xterm-zerolag-input.js',
+  ];
+  const manifest = {};
+  for (const file of HASHABLE) {
+    const filePath = join(distPublic, file);
+    const content = readFileSync(filePath);
+    const hash = createHash('md5').update(content).digest('hex').slice(0, 8);
+    const ext = extname(file);
+    const base = basename(file, ext);
+    const dir = dirname(file);
+    const hashed = dir === '.' ? `${base}.${hash}${ext}` : `${dir}/${base}.${hash}${ext}`;
+    renameSync(filePath, join(distPublic, hashed));
+    manifest[file] = hashed;
+  }
+  // Rewrite index.html to reference hashed filenames
+  let html = readFileSync(join(distPublic, 'index.html'), 'utf8');
+  for (const [original, hashed] of Object.entries(manifest)) {
+    html = html.replaceAll(`"${original}"`, `"${hashed}"`);
+  }
+  writeFileSync(join(distPublic, 'index.html'), html);
+  console.log('  Hashed files:');
+  for (const [orig, hashed] of Object.entries(manifest)) {
+    console.log(`    ${orig} -> ${hashed}`);
+  }
+}
+
+// 6. Compress with gzip + brotli
 run(
   'compress',
   `for f in dist/web/public/*.js dist/web/public/*.css dist/web/public/*.html dist/web/public/vendor/*.js dist/web/public/vendor/*.css; do` +

scripts/capture-readme-screenshots.mjs

@@ -428,7 +428,7 @@ const SUBAGENT_ACTIVITY = {
   'agent-002': [
     { type: 'tool', tool: 'Glob', input: { pattern: 'test/**/*.test.ts' }, timestamp: new Date().toISOString(), agentId: 'agent-002' },
     { type: 'tool', tool: 'Read', input: { file_path: '/home/arkon/codeman/test/respawn-test-utils.ts' }, timestamp: new Date().toISOString(), agentId: 'agent-002' },
-    { type: 'tool', tool: 'Read', input: { file_path: '/home/arkon/codeman/vitest.config.ts' }, timestamp: new Date().toISOString(), agentId: 'agent-002' },
+    { type: 'tool', tool: 'Read', input: { file_path: '/home/arkon/codeman/config/vitest.config.ts' }, timestamp: new Date().toISOString(), agentId: 'agent-002' },
     { type: 'message', role: 'assistant', text: 'Analyzing test patterns: MockSession, unique ports, fileParallelism: false...', timestamp: new Date().toISOString(), agentId: 'agent-002' },
   ],
 };

scripts/capture-video-screenshots.mjs

@@ -9,7 +9,7 @@
  *
  * Usage:  node scripts/capture-video-screenshots.mjs
  * Port:   3198 (static file server)
- * Output: remotion/public/ (6 PNGs)
+ * Output: scripts/scripts/remotion/public/ (6 PNGs)
  */
 
 import { chromium } from 'playwright';
@@ -21,7 +21,7 @@ import { fileURLToPath } from 'url';
 const __dirname = fileURLToPath(new URL('.', import.meta.url));
 const PROJECT_ROOT = join(__dirname, '..');
 const PUBLIC_DIR = join(PROJECT_ROOT, 'src', 'web', 'public');
-const OUTPUT_DIR = join(PROJECT_ROOT, 'remotion', 'public');
+const OUTPUT_DIR = join(PROJECT_ROOT, 'scripts', 'remotion', 'public');
 const PORT = 3198;
 
 const DESKTOP_VIEWPORT = { width: 1920, height: 1080 };
@@ -516,7 +516,7 @@ async function captureDesktopWelcome(browser) {
       path: join(OUTPUT_DIR, 'desktop-welcome.png'),
       fullPage: false,
     });
-    console.log('  Saved: remotion/public/desktop-welcome.png');
+    console.log('  Saved: scripts/remotion/public/desktop-welcome.png');
   } finally {
     await context.close();
   }
@@ -545,7 +545,7 @@ async function captureDesktopClaude(browser) {
       path: join(OUTPUT_DIR, 'desktop-claude.png'),
       fullPage: false,
     });
-    console.log('  Saved: remotion/public/desktop-claude.png');
+    console.log('  Saved: scripts/remotion/public/desktop-claude.png');
   } finally {
     await context.close();
   }
@@ -575,7 +575,7 @@ async function captureDesktopBothClaude(browser) {
       path: join(OUTPUT_DIR, 'desktop-both-claude.png'),
       fullPage: false,
     });
-    console.log('  Saved: remotion/public/desktop-both-claude.png');
+    console.log('  Saved: scripts/remotion/public/desktop-both-claude.png');
   } finally {
     await context.close();
   }
@@ -605,7 +605,7 @@ async function captureDesktopBothOpencode(browser) {
       path: join(OUTPUT_DIR, 'desktop-both-opencode.png'),
       fullPage: false,
     });
-    console.log('  Saved: remotion/public/desktop-both-opencode.png');
+    console.log('  Saved: scripts/remotion/public/desktop-both-opencode.png');
   } finally {
     await context.close();
   }
@@ -634,7 +634,7 @@ async function captureMobileClaude(browser) {
       path: join(OUTPUT_DIR, 'mobile-claude.png'),
       fullPage: false,
     });
-    console.log('  Saved: remotion/public/mobile-claude.png');
+    console.log('  Saved: scripts/remotion/public/mobile-claude.png');
   } finally {
     await context.close();
   }
@@ -663,7 +663,7 @@ async function captureMobileOpencode(browser) {
       path: join(OUTPUT_DIR, 'mobile-opencode.png'),
       fullPage: false,
     });
-    console.log('  Saved: remotion/public/mobile-opencode.png');
+    console.log('  Saved: scripts/remotion/public/mobile-opencode.png');
   } finally {
     await context.close();
   }
@@ -706,12 +706,12 @@ async function main() {
     console.log('All 6 screenshots captured!');
     console.log('='.repeat(60));
     console.log('\nOutput files:');
-    console.log('  remotion/public/desktop-welcome.png');
-    console.log('  remotion/public/desktop-claude.png');
-    console.log('  remotion/public/desktop-both-claude.png');
-    console.log('  remotion/public/desktop-both-opencode.png');
-    console.log('  remotion/public/mobile-claude.png');
-    console.log('  remotion/public/mobile-opencode.png');
+    console.log('  scripts/remotion/public/desktop-welcome.png');
+    console.log('  scripts/remotion/public/desktop-claude.png');
+    console.log('  scripts/remotion/public/desktop-both-claude.png');
+    console.log('  scripts/remotion/public/desktop-both-opencode.png');
+    console.log('  scripts/remotion/public/mobile-claude.png');
+    console.log('  scripts/remotion/public/mobile-opencode.png');
   } catch (err) {
     console.error('\nFatal error:', err.message);
     console.error(err.stack);

scripts/codeman-tunnel-named.service

@@ -0,0 +1,19 @@
+[Unit]
+Description=Codeman Cloudflare Named Tunnel
+After=network-online.target codeman-web.service
+Wants=network-online.target
+
+[Service]
+Type=simple
+ExecStart=/usr/bin/cloudflared tunnel --config %h/.cloudflared/codeman.yml run codeman
+Restart=always
+RestartSec=5
+KillMode=process
+
+# Logging
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=codeman-tunnel-named
+
+[Install]
+WantedBy=default.target

scripts/codeman-web.service

@@ -11,6 +11,7 @@ RestartSec=5
 KillMode=process
 Environment=NODE_ENV=production
 Environment=HOME=/home/arkon
+Environment=NODE_COMPILE_CACHE=/home/arkon/.codeman/compile-cache
 
 # Logging
 StandardOutput=journal

scripts/postinstall.js

@@ -250,10 +250,10 @@ if (isGlobalInstall) {
 } else {
     try {
         const require = createRequire(import.meta.url);
-        const xtermDir = join(require.resolve('xterm'), '..', '..');
-        const fitDir = join(require.resolve('xterm-addon-fit'), '..', '..');
-        const webglDir = join(require.resolve('xterm-addon-webgl'), '..', '..');
-        const unicode11Dir = join(require.resolve('xterm-addon-unicode11'), '..', '..');
+        const xtermDir = join(require.resolve('@xterm/xterm'), '..', '..');
+        const fitDir = join(require.resolve('@xterm/addon-fit'), '..', '..');
+        const webglDir = join(require.resolve('@xterm/addon-webgl'), '..', '..');
+        const unicode11Dir = join(require.resolve('@xterm/addon-unicode11'), '..', '..');
         const vendorDir = join(srcDir, 'web', 'public', 'vendor');
 
         const { mkdirSync, copyFileSync } = await import('fs');
@@ -263,19 +263,47 @@ if (isGlobalInstall) {
         // Minify xterm JS for dev vendor dir (npm packages don't ship .min.js)
         try {
             execSync(`npx esbuild "${join(xtermDir, 'lib', 'xterm.js')}" --minify --outfile="${join(vendorDir, 'xterm.min.js')}"`, { stdio: 'pipe' });
-            execSync(`npx esbuild "${join(fitDir, 'lib', 'xterm-addon-fit.js')}" --minify --outfile="${join(vendorDir, 'xterm-addon-fit.min.js')}"`, { stdio: 'pipe' });
-            execSync(`npx esbuild "${join(unicode11Dir, 'lib', 'xterm-addon-unicode11.js')}" --minify --outfile="${join(vendorDir, 'xterm-addon-unicode11.min.js')}"`, { stdio: 'pipe' });
+            execSync(`npx esbuild "${join(fitDir, 'lib', 'addon-fit.js')}" --minify --outfile="${join(vendorDir, 'xterm-addon-fit.min.js')}"`, { stdio: 'pipe' });
+            execSync(`npx esbuild "${join(unicode11Dir, 'lib', 'addon-unicode11.js')}" --minify --outfile="${join(vendorDir, 'xterm-addon-unicode11.min.js')}"`, { stdio: 'pipe' });
             console.log(colors.green('✓ xterm vendor files copied to src/web/public/vendor/'));
         } catch {
             // Fallback: copy unminified
             copyFileSync(join(xtermDir, 'lib', 'xterm.js'), join(vendorDir, 'xterm.min.js'));
-            copyFileSync(join(fitDir, 'lib', 'xterm-addon-fit.js'), join(vendorDir, 'xterm-addon-fit.min.js'));
-            copyFileSync(join(unicode11Dir, 'lib', 'xterm-addon-unicode11.js'), join(vendorDir, 'xterm-addon-unicode11.min.js'));
+            copyFileSync(join(fitDir, 'lib', 'addon-fit.js'), join(vendorDir, 'xterm-addon-fit.min.js'));
+            copyFileSync(join(unicode11Dir, 'lib', 'addon-unicode11.js'), join(vendorDir, 'xterm-addon-unicode11.min.js'));
             console.log(colors.green('✓ xterm vendor files copied') + colors.dim(' (unminified — esbuild not available)'));
         }
 
         // WebGL addon: copy unminified (matches build script behavior)
-        copyFileSync(join(webglDir, 'lib', 'xterm-addon-webgl.js'), join(vendorDir, 'xterm-addon-webgl.min.js'));
+        copyFileSync(join(webglDir, 'lib', 'addon-webgl.js'), join(vendorDir, 'xterm-addon-webgl.min.js'));
+
+        // xterm-zerolag-input: bundle local package as IIFE for <script> tag loading
+        try {
+            const zerolagSrc = join(import.meta.dirname, '..', 'packages', 'xterm-zerolag-input', 'src', 'zerolag-input-addon.ts');
+            const zerolagOut = join(vendorDir, 'xterm-zerolag-input.js');
+            execSync(
+                `npx esbuild "${zerolagSrc}" --bundle --format=iife --global-name=XtermZerolagInput --outfile="${zerolagOut}"`,
+                { stdio: 'pipe' }
+            );
+            // Append global aliases so app.js can use `new LocalEchoOverlay(terminal)`
+            const { appendFileSync } = await import('fs');
+            appendFileSync(
+                zerolagOut,
+                '\n// Global aliases for browser usage\n' +
+                'if(typeof window!=="undefined"){' +
+                    'window.ZerolagInputAddon=XtermZerolagInput.ZerolagInputAddon;' +
+                    'window.LocalEchoOverlay=class extends XtermZerolagInput.ZerolagInputAddon{' +
+                        'constructor(terminal){' +
+                            'super({prompt:{type:"character",char:"\\u276f",offset:2}});' +
+                            'this.activate(terminal);' +
+                        '}' +
+                    '};' +
+                '}\n'
+            );
+            console.log(colors.green('✓ xterm-zerolag-input bundled to vendor/'));
+        } catch {
+            console.log(colors.yellow('⚠ Failed to bundle xterm-zerolag-input — overlay may not work in dev mode'));
+        }
     } catch (err) {
         hasWarnings = true;
         console.log(colors.yellow('⚠ Failed to copy xterm vendor files'));
@@ -284,6 +312,36 @@ if (isGlobalInstall) {
     }
 }
 
+// ----------------------------------------------------------------------------
+// 5. Install git pre-commit hook (format check)
+// ----------------------------------------------------------------------------
+
+if (!isGlobalInstall) {
+    try {
+        const { writeFileSync, mkdirSync } = await import('fs');
+        const gitHooksDir = join(import.meta.dirname, '..', '.git', 'hooks');
+        if (existsSync(join(import.meta.dirname, '..', '.git'))) {
+            mkdirSync(gitHooksDir, { recursive: true });
+            const hook = `#!/bin/bash
+# Auto-installed by postinstall — prevents CI format failures
+staged_ts=$(git diff --cached --name-only --diff-filter=ACM -- '*.ts')
+[ -z "$staged_ts" ] && exit 0
+echo "$staged_ts" | xargs npx prettier --check 2>&1
+if [ $? -ne 0 ]; then
+    echo ""
+    echo "Pre-commit: Prettier check failed. Run 'npm run format' to fix."
+    exit 1
+fi
+`;
+            const hookPath = join(gitHooksDir, 'pre-commit');
+            writeFileSync(hookPath, hook, { mode: 0o755 });
+            console.log(colors.green('✓ Git pre-commit hook installed (prettier check)'));
+        }
+    } catch {
+        // Non-critical — git hook is a convenience
+    }
+}
+
 // ----------------------------------------------------------------------------
 // Summary
 // ----------------------------------------------------------------------------

scripts/remotion/Root.tsx

@@ -0,0 +1,27 @@
+import React from 'react';
+import { Composition } from 'remotion';
+import { CodemanDemo, TOTAL_FRAMES } from './compositions/CodemanDemo';
+import { ZerolagDemo, ZEROLAG_TOTAL_FRAMES } from './compositions/ZerolagDemo';
+
+export const RemotionRoot: React.FC = () => {
+  return (
+    <>
+      <Composition
+        id="CodemanDemo"
+        component={CodemanDemo}
+        durationInFrames={TOTAL_FRAMES}
+        fps={30}
+        width={1920}
+        height={1080}
+      />
+      <Composition
+        id="ZerolagDemo"
+        component={ZerolagDemo}
+        durationInFrames={ZEROLAG_TOTAL_FRAMES}
+        fps={30}
+        width={1920}
+        height={1080}
+      />
+    </>
+  );
+};

scripts/remotion/components/IOSKeyboard.tsx

@@ -0,0 +1,207 @@
+import React from 'react';
+import { fonts } from '../lib/theme';
+
+type IOSKeyboardProps = {
+  activeKey?: string;
+  /** How many frames since the key was pressed (for highlight decay) */
+  pressAge?: number;
+};
+
+const ROW_1 = ['q', 'w', 'e', 'r', 't', 'y', 'u', 'i', 'o', 'p'];
+const ROW_2 = ['a', 's', 'd', 'f', 'g', 'h', 'j', 'k', 'l'];
+const ROW_3 = ['z', 'x', 'c', 'v', 'b', 'n', 'm'];
+
+const KEY_H = 42;
+const KEY_GAP = 6;
+const ROW_GAP = 11;
+const SIDE_PAD = 3;
+
+const KEYBOARD_BG = '#1c1c1e';
+const KEY_BG = '#3a3a3c';
+const KEY_BG_ACTIVE = '#636366';
+const KEY_TEXT = '#fff';
+const SPECIAL_BG = '#2c2c2e';
+
+const Key: React.FC<{
+  label: string;
+  width: number;
+  isActive: boolean;
+  pressAge: number;
+  fontSize?: number;
+}> = ({ label, width, isActive, pressAge, fontSize = 22 }) => {
+  // Highlight decays over 4 frames
+  const highlightOpacity = isActive && pressAge < 4 ? 1 - pressAge / 4 : 0;
+  const bg = highlightOpacity > 0
+    ? lerpColor(KEY_BG, KEY_BG_ACTIVE, highlightOpacity)
+    : KEY_BG;
+
+  return (
+    <div
+      style={{
+        width,
+        height: KEY_H,
+        borderRadius: 5,
+        background: bg,
+        display: 'flex',
+        justifyContent: 'center',
+        alignItems: 'center',
+        fontSize,
+        fontFamily: fonts.ui,
+        color: KEY_TEXT,
+        fontWeight: 300,
+        flexShrink: 0,
+      }}
+    >
+      {label}
+    </div>
+  );
+};
+
+function lerpColor(a: string, b: string, t: number): string {
+  const pa = parseInt(a.slice(1), 16);
+  const pb = parseInt(b.slice(1), 16);
+  const r = Math.round(((pa >> 16) & 0xff) * (1 - t) + ((pb >> 16) & 0xff) * t);
+  const g = Math.round(((pa >> 8) & 0xff) * (1 - t) + ((pb >> 8) & 0xff) * t);
+  const bl = Math.round((pa & 0xff) * (1 - t) + (pb & 0xff) * t);
+  return `#${((r << 16) | (g << 8) | bl).toString(16).padStart(6, '0')}`;
+}
+
+export const IOSKeyboard: React.FC<IOSKeyboardProps> = ({ activeKey, pressAge = 99 }) => {
+  const isActive = (key: string) =>
+    activeKey !== undefined && key.toLowerCase() === activeKey.toLowerCase();
+  const age = (key: string) => (isActive(key) ? pressAge : 99);
+
+  // Backspace/delete key highlight
+  const deleteHighlight = activeKey === '⌫' && pressAge < 4 ? 1 - pressAge / 4 : 0;
+  const deleteBg = deleteHighlight > 0 ? lerpColor(SPECIAL_BG, KEY_BG_ACTIVE, deleteHighlight) : SPECIAL_BG;
+
+  // Key widths: 10 keys + 9 gaps in ~375px row → each key ~33px
+  const letterKeyW = 33;
+  // Row 2 has 9 keys → same key width but centered with side padding
+  // Row 3 has shift + 7 keys + delete
+
+  return (
+    <div
+      style={{
+        width: '100%',
+        background: KEYBOARD_BG,
+        padding: `${ROW_GAP}px ${SIDE_PAD}px 20px`,
+        display: 'flex',
+        flexDirection: 'column',
+        gap: ROW_GAP,
+      }}
+    >
+      {/* Row 1: q-p */}
+      <div style={{ display: 'flex', gap: KEY_GAP, justifyContent: 'center' }}>
+        {ROW_1.map((k) => (
+          <Key key={k} label={k} width={letterKeyW} isActive={isActive(k)} pressAge={age(k)} />
+        ))}
+      </div>
+
+      {/* Row 2: a-l */}
+      <div style={{ display: 'flex', gap: KEY_GAP, justifyContent: 'center' }}>
+        {ROW_2.map((k) => (
+          <Key key={k} label={k} width={letterKeyW} isActive={isActive(k)} pressAge={age(k)} />
+        ))}
+      </div>
+
+      {/* Row 3: shift + z-m + delete */}
+      <div style={{ display: 'flex', gap: KEY_GAP, justifyContent: 'center' }}>
+        <div
+          style={{
+            width: 42,
+            height: KEY_H,
+            borderRadius: 5,
+            background: SPECIAL_BG,
+            display: 'flex',
+            justifyContent: 'center',
+            alignItems: 'center',
+          }}
+        >
+          <svg width="20" height="16" viewBox="0 0 20 16" fill="none">
+            <path d="M10 2L17 9H13V14H7V9H3L10 2Z" fill="#fff" />
+          </svg>
+        </div>
+        {ROW_3.map((k) => (
+          <Key key={k} label={k} width={letterKeyW} isActive={isActive(k)} pressAge={age(k)} />
+        ))}
+        <div
+          style={{
+            width: 42,
+            height: KEY_H,
+            borderRadius: 5,
+            background: deleteBg,
+            display: 'flex',
+            justifyContent: 'center',
+            alignItems: 'center',
+          }}
+        >
+          <svg width="22" height="16" viewBox="0 0 22 16" fill="none">
+            <path d="M7 1L1 8L7 15H21V1H7Z" stroke="#fff" strokeWidth="1.5" fill="none" />
+            <path d="M12 5L17 10M17 5L12 10" stroke="#fff" strokeWidth="1.5" strokeLinecap="round" />
+          </svg>
+        </div>
+      </div>
+
+      {/* Row 4: 123 / globe / space / return */}
+      <div style={{ display: 'flex', gap: KEY_GAP, justifyContent: 'center' }}>
+        <div
+          style={{
+            width: 42,
+            height: KEY_H,
+            borderRadius: 5,
+            background: SPECIAL_BG,
+            display: 'flex',
+            justifyContent: 'center',
+            alignItems: 'center',
+            fontSize: 15,
+            fontFamily: fonts.ui,
+            color: '#fff',
+          }}
+        >
+          123
+        </div>
+        <div
+          style={{
+            width: 38,
+            height: KEY_H,
+            borderRadius: 5,
+            background: SPECIAL_BG,
+            display: 'flex',
+            justifyContent: 'center',
+            alignItems: 'center',
+          }}
+        >
+          <svg width="20" height="20" viewBox="0 0 20 20" fill="none">
+            <circle cx="10" cy="10" r="8" stroke="#fff" strokeWidth="1.2" />
+            <path d="M4 10H16M10 4C7 7 7 13 10 16M10 4C13 7 13 13 10 16" stroke="#fff" strokeWidth="1" />
+          </svg>
+        </div>
+        {/* Space bar */}
+        <Key
+          label="space"
+          width={186}
+          isActive={isActive(' ')}
+          pressAge={age(' ')}
+          fontSize={15}
+        />
+        <div
+          style={{
+            width: 88,
+            height: KEY_H,
+            borderRadius: 5,
+            background: SPECIAL_BG,
+            display: 'flex',
+            justifyContent: 'center',
+            alignItems: 'center',
+            fontSize: 15,
+            fontFamily: fonts.ui,
+            color: '#fff',
+          }}
+        >
+          return
+        </div>
+      </div>
+    </div>
+  );
+};

scripts/remotion/components/IPhone17ProFrame.tsx

@@ -0,0 +1,171 @@
+import React from 'react';
+import { spring, useCurrentFrame, useVideoConfig } from 'remotion';
+
+/**
+ * Pixel-accurate iPhone 17 Pro frame.
+ *
+ * Dimensions based on iPhone 16 Pro (same form factor):
+ * - Screen: 393×852 CSS points (2622×1206 @3x)
+ * - Corner radius: 55px (device), 50px (screen inner)
+ * - Bezel: ~3.5px (thinnest in any iPhone)
+ * - Dynamic Island: 126×37 pill, centered 13px from top
+ * - Safe area: top 59px, bottom 34px (home indicator)
+ * - Frame: natural titanium (#8a8a8e border)
+ */
+
+type IPhone17ProFrameProps = {
+  children: React.ReactNode;
+  /** Disable the spring entrance animation */
+  noAnimation?: boolean;
+};
+
+// Device dimensions (CSS points)
+const SCREEN_W = 393;
+const SCREEN_H = 852;
+const BEZEL = 4;
+const DEVICE_W = SCREEN_W + BEZEL * 2; // 401
+const DEVICE_H = SCREEN_H + BEZEL * 2; // 860
+const DEVICE_RADIUS = 55;
+const SCREEN_RADIUS = 50;
+
+// Dynamic Island
+const DI_W = 126;
+const DI_H = 37;
+const DI_TOP = 13; // from top of screen
+const DI_RADIUS = DI_H / 2; // pill shape
+
+export { SCREEN_W, SCREEN_H };
+
+export const IPhone17ProFrame: React.FC<IPhone17ProFrameProps> = ({ children, noAnimation }) => {
+  const frame = useCurrentFrame();
+  const { fps } = useVideoConfig();
+
+  const scale = noAnimation
+    ? 1
+    : spring({ frame, fps, config: { damping: 15, stiffness: 80 } });
+
+  return (
+    <div
+      style={{
+        width: DEVICE_W,
+        height: DEVICE_H,
+        transform: `scale(${scale})`,
+        transformOrigin: 'center center',
+        position: 'relative',
+      }}
+    >
+      {/* Titanium frame (outer body) */}
+      <div
+        style={{
+          width: DEVICE_W,
+          height: DEVICE_H,
+          borderRadius: DEVICE_RADIUS,
+          background: '#2c2c2e', // dark titanium
+          border: '1.5px solid #48484a', // subtle edge highlight
+          boxShadow: [
+            '0 2px 4px rgba(0,0,0,0.3)', // close shadow
+            '0 12px 40px rgba(0,0,0,0.5)', // mid shadow
+            '0 30px 80px rgba(0,0,0,0.4)', // far shadow
+            'inset 0 1px 0 rgba(255,255,255,0.05)', // top edge gleam
+          ].join(', '),
+          position: 'relative',
+          overflow: 'hidden',
+        }}
+      >
+        {/* Side button (right) — power */}
+        <div
+          style={{
+            position: 'absolute',
+            right: -2,
+            top: 180,
+            width: 3,
+            height: 65,
+            borderRadius: '0 2px 2px 0',
+            background: '#48484a',
+          }}
+        />
+
+        {/* Side buttons (left) — volume up, down, action */}
+        <div
+          style={{
+            position: 'absolute',
+            left: -2,
+            top: 140,
+            width: 3,
+            height: 28,
+            borderRadius: '2px 0 0 2px',
+            background: '#48484a',
+          }}
+        />
+        <div
+          style={{
+            position: 'absolute',
+            left: -2,
+            top: 185,
+            width: 3,
+            height: 50,
+            borderRadius: '2px 0 0 2px',
+            background: '#48484a',
+          }}
+        />
+        <div
+          style={{
+            position: 'absolute',
+            left: -2,
+            top: 250,
+            width: 3,
+            height: 50,
+            borderRadius: '2px 0 0 2px',
+            background: '#48484a',
+          }}
+        />
+
+        {/* Screen */}
+        <div
+          style={{
+            position: 'absolute',
+            top: BEZEL,
+            left: BEZEL,
+            width: SCREEN_W,
+            height: SCREEN_H,
+            borderRadius: SCREEN_RADIUS,
+            overflow: 'hidden',
+            background: '#000',
+          }}
+        >
+          {/* App content */}
+          {children}
+
+          {/* Dynamic Island (on top of everything) */}
+          <div
+            style={{
+              position: 'absolute',
+              top: DI_TOP,
+              left: (SCREEN_W - DI_W) / 2,
+              width: DI_W,
+              height: DI_H,
+              borderRadius: DI_RADIUS,
+              background: '#000',
+              zIndex: 50,
+            }}
+          />
+        </div>
+      </div>
+
+      {/* Home indicator */}
+      <div
+        style={{
+          position: 'absolute',
+          bottom: BEZEL + 8,
+          left: '50%',
+          transform: 'translateX(-50%)',
+          width: 134,
+          height: 5,
+          borderRadius: 3,
+          background: 'rgba(255,255,255,0.2)',
+          zIndex: 60,
+        }}
+      />
+    </div>
+  );
+};

scripts/remotion/components/TerminalScreen.tsx

@@ -0,0 +1,117 @@
+import React from 'react';
+import { useCurrentFrame } from 'remotion';
+import { colors, fonts } from '../lib/theme';
+
+type OverlayChar = {
+  char: string;
+  confirmed: boolean;
+};
+
+type TerminalScreenProps = {
+  typed: string;
+  cursorVisible: boolean;
+  overlayChars?: OverlayChar[];
+  fontSize?: number;
+};
+
+/**
+ * Renders a terminal area styled exactly like the real Codeman xterm.js terminal.
+ * Background #0d0d0d, Fira Code font, block cursor, Claude Code prompt.
+ */
+export const TerminalScreen: React.FC<TerminalScreenProps> = ({
+  typed,
+  cursorVisible,
+  overlayChars,
+  fontSize = 28,
+}) => {
+  const frame = useCurrentFrame();
+
+  // Block cursor — solid, no blink (matches Codeman's cursorBlink: false)
+  const cursorOn = cursorVisible;
+  // But add a subtle blink for video clarity so viewers notice it
+  const cursorOpacity = cursorOn ? (Math.floor(frame / 20) % 2 === 0 ? 0.9 : 0.6) : 0;
+
+  const lineHeight = Math.round(fontSize * 1.35);
+  const charWidth = fontSize * 0.6;
+
+  return (
+    <div
+      style={{
+        width: '100%',
+        height: '100%',
+        background: '#0d0d0d',
+        position: 'relative',
+        fontFamily: '"Fira Code", "Cascadia Code", "JetBrains Mono", "SF Mono", Monaco, monospace',
+        overflow: 'hidden',
+      }}
+    >
+      {/* Previous terminal output lines (fake history for realism) */}
+      <div
+        style={{
+          padding: '16px 20px',
+          fontSize: fontSize * 0.65,
+          lineHeight: `${Math.round(fontSize * 0.65 * 1.4)}px`,
+          color: '#495057',
+        }}
+      >
+        <div>
+          <span style={{ color: '#339af0' }}>❯</span>
+          <span style={{ color: '#495057' }}> claude --dangerously-skip-permissions</span>
+        </div>
+        <div style={{ color: '#3a3a3a', marginTop: 4 }}>
+          ╭────────────────────────────────────╮
+        </div>
+        <div style={{ color: '#3a3a3a' }}>
+          │ Claude Code session active         │
+        </div>
+        <div style={{ color: '#3a3a3a' }}>
+          ╰────────────────────────────────────╯
+        </div>
+      </div>
+
+      {/* Active prompt line — this is where the typing happens */}
+      <div
+        style={{
+          padding: '8px 20px',
+          fontSize,
+          lineHeight: `${lineHeight}px`,
+          display: 'flex',
+          alignItems: 'baseline',
+        }}
+      >
+        {/* Prompt character */}
+        <span style={{ color: '#339af0', fontWeight: 700, marginRight: charWidth * 0.8 }}>❯</span>
+
+        {/* Typed text */}
+        {overlayChars ? (
+          // Zerolag mode: show overlay chars with confirmed/unconfirmed color
+          overlayChars.map((oc, i) => (
+            <span
+              key={i}
+              style={{
+                color: oc.confirmed ? '#e0e0e0' : '#7a7a7a',
+                letterSpacing: '0.5px',
+              }}
+            >
+              {oc.char}
+            </span>
+          ))
+        ) : (
+          <span style={{ color: '#e0e0e0', letterSpacing: '0.5px' }}>{typed}</span>
+        )}
+
+        {/* Block cursor */}
+        <span
+          style={{
+            display: 'inline-block',
+            width: charWidth,
+            height: lineHeight * 0.85,
+            background: `rgba(224, 224, 224, ${cursorOpacity})`,
+            verticalAlign: 'text-bottom',
+            marginLeft: 1,
+          }}
+        />
+      </div>
+    </div>
+  );
+};

scripts/remotion/compositions/ZerolagDemo.tsx

@@ -0,0 +1,569 @@
+import React from 'react';
+import {
+  AbsoluteFill,
+  Img,
+  interpolate,
+  Sequence,
+  spring,
+  staticFile,
+  useCurrentFrame,
+  useVideoConfig,
+} from 'remotion';
+import { colors, fonts } from '../lib/theme';
+import { IPhone17ProFrame, SCREEN_W, SCREEN_H } from '../components/IPhone17ProFrame';
+import { IOSKeyboard } from '../components/IOSKeyboard';
+
+// ─── Scene timing (frames @ 30fps) ───
+const TITLE_DUR = 60;
+const PHONES_DUR = 30;
+const TYPING_DUR = 610;
+const HOLD_DUR = 50;
+const OUTRO_DUR = 45;
+
+const TITLE_START = 0;
+const PHONES_START = TITLE_DUR; // 60
+const TYPING_START = PHONES_START + PHONES_DUR; // 90
+const HOLD_START = TYPING_START + TYPING_DUR; // 700
+const OUTRO_START = HOLD_START + HOLD_DUR; // 750
+
+export const ZEROLAG_TOTAL_FRAMES = OUTRO_START + OUTRO_DUR; // 795
+
+// ─── iPhone 17 Pro safe area ───
+const SAFE_AREA_TOP = 59; // Below Dynamic Island
+const PHONE_SCALE = 1.12; // Scale up to fill more of the frame
+
+// Claude Code header: tab bar + session info + prompt context from screenshot
+const HEADER_H = 120;
+
+// Terminal typing overlay: aligned with the ❯ prompt position in the Claude Code screenshot
+const TERMINAL_TOP = 185;
+const TERMINAL_LEFT = 14;
+const TERMINAL_FONT = 21; // Slightly smaller to fit toolbar below
+
+// Codeman toolbar from screenshot (bottom section showing /init, /clear, Run, etc.)
+const TOOLBAR_H = 95;
+
+// ─── Typing schedule ───
+const CORRECT_TEXT =
+  'zerolag technology brings in a visual dom overlay to make typing instant, even if your codeman server is on the other side of the world';
+const FRAME_GAP = 4; // ~133ms per keystroke (~75 WPM)
+
+// Remote connection lag: 600ms–2.7s per char (18–80 frames @ 30fps).
+// Periodic spikes simulate packet loss / retransmission bursts.
+// TCP head-of-line blocking causes a single spike to freeze all subsequent chars.
+const LAGGY_DELAYS = [
+  24, 30, 26, 32, 72, 28, 22, 34, 26, 30, 20, 28, 36, 24, 30, 22, 26, 34, 28, 20, 68, 30, 24, 32,
+  26, 22, 28, 34, 30, 26, 32, 24, 80, 22, 30, 26, 28, 34, 24, 30,
+];
+
+type KeyAction = { frame: number; char: string; lagDelay: number };
+
+const buildSchedule = (): KeyAction[] => {
+  const actions: KeyAction[] = [];
+  const lag = (i: number) => LAGGY_DELAYS[i % LAGGY_DELAYS.length];
+
+  for (let i = 0; i < CORRECT_TEXT.length; i++) {
+    actions.push({ frame: i * FRAME_GAP, char: CORRECT_TEXT[i], lagDelay: lag(i) });
+  }
+
+  return actions;
+};
+
+const TYPING_SCHEDULE = buildSchedule();
+
+/**
+ * Replay actions in order up to current frame, computing the visible text buffer.
+ * TCP-ordered: stops at first unresolved echo (head-of-line blocking).
+ */
+const computeVisibleText = (frame: number, withLag: boolean): string => {
+  let buffer = '';
+  for (const a of TYPING_SCHEDULE) {
+    const threshold = withLag ? a.frame + a.lagDelay : a.frame;
+    if (frame < threshold) break;
+    buffer += a.char;
+  }
+  return buffer;
+};
+
+// ─── iOS Status Bar (sits in the safe area, flanking Dynamic Island) ───
+const IOSStatusBar: React.FC = () => (
+  <div
+    style={{
+      position: 'absolute',
+      top: 17,
+      left: 0,
+      right: 0,
+      height: 20,
+      display: 'flex',
+      justifyContent: 'space-between',
+      padding: '0 30px',
+      fontSize: 15,
+      fontFamily: fonts.ui,
+      fontWeight: 600,
+      color: '#fff',
+      zIndex: 40,
+    }}
+  >
+    <span>9:41</span>
+    <div style={{ display: 'flex', gap: 6, alignItems: 'center' }}>
+      {/* Signal */}
+      <svg width="17" height="12" viewBox="0 0 17 12">
+        <rect x="0" y="8" width="3" height="4" rx="0.5" fill="#fff" />
+        <rect x="4.5" y="5" width="3" height="7" rx="0.5" fill="#fff" />
+        <rect x="9" y="2" width="3" height="10" rx="0.5" fill="#fff" />
+        <rect x="13.5" y="0" width="3" height="12" rx="0.5" fill="#fff" />
+      </svg>
+      {/* WiFi */}
+      <svg width="16" height="12" viewBox="0 0 16 12">
+        <path
+          d="M4.5 8.5C5.5 7.2 6.7 6.5 8 6.5s2.5.7 3.5 2"
+          stroke="#fff"
+          strokeWidth="1.5"
+          fill="none"
+          strokeLinecap="round"
+        />
+        <path
+          d="M1.5 5.5C3.5 3 5.7 1.5 8 1.5s4.5 1.5 6.5 4"
+          stroke="#fff"
+          strokeWidth="1.5"
+          fill="none"
+          strokeLinecap="round"
+        />
+        <circle cx="8" cy="11" r="1.5" fill="#fff" />
+      </svg>
+      {/* Battery */}
+      <svg width="27" height="12" viewBox="0 0 27 12">
+        <rect x="0" y="0.5" width="23" height="11" rx="2" stroke="#fff" strokeWidth="1" fill="none" />
+        <rect x="24" y="3.5" width="2.5" height="5" rx="1" fill="#fff" opacity="0.4" />
+        <rect x="1.5" y="2" width="20" height="8" rx="1" fill="#32d74b" />
+      </svg>
+    </div>
+  </div>
+);
+
+// ─── Typing overlay ───
+const TypingOverlay: React.FC<{
+  typed: string;
+  cursorVisible: boolean;
+}> = ({ typed, cursorVisible }) => {
+  const frame = useCurrentFrame();
+  const cursorOpacity = cursorVisible ? (Math.floor(frame / 18) % 2 === 0 ? 0.85 : 0.5) : 0;
+  const lineH = Math.round(TERMINAL_FONT * 1.4);
+  const charW = TERMINAL_FONT * 0.62;
+
+  return (
+    <div
+      style={{
+        position: 'absolute',
+        top: TERMINAL_TOP,
+        left: TERMINAL_LEFT,
+        right: TERMINAL_LEFT,
+        fontFamily: '"Fira Code", "Cascadia Code", "JetBrains Mono", "SF Mono", Monaco, monospace',
+        fontSize: TERMINAL_FONT,
+        lineHeight: `${lineH}px`,
+        zIndex: 10,
+      }}
+    >
+      <span style={{ color: '#339af0', fontWeight: 700 }}>{'❯ '}</span>
+      <span style={{ color: '#e0e0e0' }}>{typed}</span>
+      <span
+        style={{
+          display: 'inline-block',
+          width: charW,
+          height: lineH * 0.82,
+          background: `rgba(224, 224, 224, ${cursorOpacity})`,
+          verticalAlign: 'text-bottom',
+          marginLeft: 1,
+        }}
+      />
+    </div>
+  );
+};
+
+// ─── Single phone: iPhone 17 Pro + Claude Code header + typing + keyboard ───
+const MobileCodeman: React.FC<{
+  typed: string;
+  cursorVisible: boolean;
+  activeKey?: string;
+  pressAge?: number;
+  showKeyboard?: boolean;
+  noAnimation?: boolean;
+  showPromo?: boolean;
+}> = ({ typed, cursorVisible, activeKey, pressAge, showKeyboard = true, noAnimation, showPromo }) => (
+  <IPhone17ProFrame noAnimation={noAnimation}>
+    <div
+      style={{ width: SCREEN_W, height: SCREEN_H, position: 'relative', overflow: 'hidden', background: '#0d0d0d' }}
+    >
+      {/* iOS status bar in the safe area */}
+      <IOSStatusBar />
+
+      {/* Claude Code header from screenshot (tabs + session info) */}
+      <div
+        style={{
+          position: 'absolute',
+          top: SAFE_AREA_TOP,
+          left: 0,
+          right: 0,
+          height: HEADER_H,
+          overflow: 'hidden',
+          zIndex: 5,
+        }}
+      >
+        <Img
+          src={staticFile('mobile-claude.png')}
+          style={{
+            width: SCREEN_W,
+            objectFit: 'cover',
+            objectPosition: 'top left',
+          }}
+        />
+        {/* Fade to terminal background */}
+        <div
+          style={{
+            position: 'absolute',
+            bottom: 0,
+            left: 0,
+            right: 0,
+            height: 30,
+            background: 'linear-gradient(transparent, #0d0d0d)',
+          }}
+        />
+      </div>
+
+      {/* Dark mask: hides screenshot text below Claude Code info (e.g. "Try edit...") */}
+      <div
+        style={{
+          position: 'absolute',
+          top: SAFE_AREA_TOP + 95,
+          left: 0,
+          right: 0,
+          bottom: 0,
+          zIndex: 8,
+        }}
+      >
+        {/* Smooth gradient blend from screenshot into dark terminal */}
+        <div
+          style={{
+            height: 18,
+            background: 'linear-gradient(transparent, #0d0d0d)',
+          }}
+        />
+        <div style={{ flex: 1, background: '#0d0d0d' }} />
+      </div>
+
+      {/* Typing animation */}
+      <TypingOverlay typed={typed} cursorVisible={cursorVisible} />
+
+      {/* Codeman toolbar from screenshot (/init, /clear, /compact, Run, Run Shell, voice) */}
+      <div
+        style={{
+          position: 'absolute',
+          bottom: 232,
+          left: 0,
+          right: 0,
+          height: TOOLBAR_H,
+          overflow: 'hidden',
+          zIndex: 15,
+        }}
+      >
+        {/* Gradient blend at top */}
+        <div
+          style={{
+            position: 'absolute',
+            top: 0,
+            left: 0,
+            right: 0,
+            height: 18,
+            background: 'linear-gradient(#0d0d0d, transparent)',
+            zIndex: 1,
+          }}
+        />
+        <Img
+          src={staticFile('mobile-claude.png')}
+          style={{
+            width: SCREEN_W,
+            position: 'absolute',
+            bottom: 0,
+          }}
+        />
+      </div>
+
+      {/* Promo cards between text and toolbar (left phone only) */}
+      {showPromo && <PromoBanners />}
+
+      {/* iOS keyboard at bottom */}
+      {showKeyboard && (
+        <div style={{ position: 'absolute', bottom: 0, left: 0, right: 0, zIndex: 20 }}>
+          <IOSKeyboard activeKey={activeKey} pressAge={pressAge} />
+        </div>
+      )}
+    </div>
+  </IPhone17ProFrame>
+);
+
+// ─── Label above phone (prominent header) ───
+const PhoneLabel: React.FC<{
+  title: string;
+  detail: string;
+  dotColor: string;
+  detailColor: string;
+}> = ({ title, detail, dotColor, detailColor }) => (
+  <div style={{ textAlign: 'center', marginBottom: 16 }}>
+    <div
+      style={{
+        display: 'flex',
+        alignItems: 'center',
+        justifyContent: 'center',
+        gap: 10,
+        fontSize: 28,
+        fontWeight: 700,
+        fontFamily: fonts.ui,
+        color: '#fff',
+      }}
+    >
+      <div
+        style={{
+          width: 12,
+          height: 12,
+          borderRadius: '50%',
+          background: dotColor,
+          boxShadow: `0 0 12px ${dotColor}`,
+        }}
+      />
+      {title}
+    </div>
+    <div style={{ fontSize: 16, fontFamily: fonts.mono, color: detailColor, marginTop: 6, opacity: 0.9 }}>
+      {detail}
+    </div>
+  </div>
+);
+
+// ─── Promo banners (inside left phone, between text and keyboard) ───
+const PromoBanners: React.FC = () => (
+  <div
+    style={{
+      position: 'absolute',
+      bottom: 332,
+      left: 8,
+      right: 8,
+      display: 'flex',
+      flexDirection: 'column',
+      gap: 8,
+      zIndex: 15,
+    }}
+  >
+    <div
+      style={{
+        display: 'flex',
+        alignItems: 'center',
+        gap: 12,
+        padding: '12px 14px',
+        background: 'rgba(255,255,255,0.06)',
+        borderRadius: 12,
+        border: '1px solid rgba(255,255,255,0.1)',
+      }}
+    >
+      <svg width="30" height="30" viewBox="0 0 16 16" fill="#ccc" style={{ flexShrink: 0 }}>
+        <path d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z" />
+      </svg>
+      <div>
+        <div style={{ fontSize: 18, fontWeight: 700, fontFamily: fonts.ui, color: '#e0e0e0' }}>Codeman</div>
+        <div style={{ fontSize: 12, fontFamily: fonts.mono, color: '#888' }}>github.com/Ark0N/Codeman</div>
+      </div>
+    </div>
+    <div
+      style={{
+        display: 'flex',
+        alignItems: 'center',
+        gap: 12,
+        padding: '12px 14px',
+        background: 'rgba(255,255,255,0.06)',
+        borderRadius: 12,
+        border: '1px solid rgba(255,255,255,0.1)',
+      }}
+    >
+      <svg width="30" height="30" viewBox="0 0 16 16" style={{ flexShrink: 0 }}>
+        <rect width="16" height="16" rx="2" fill="#cb3837" />
+        <path d="M3 3h10v10H8V5H5v8H3z" fill="#fff" />
+      </svg>
+      <div>
+        <div style={{ fontSize: 18, fontWeight: 700, fontFamily: fonts.ui, color: '#e0e0e0' }}>xterm-zerolag-input</div>
+        <div style={{ fontSize: 12, fontFamily: fonts.mono, color: '#888' }}>npmjs.com/package/xterm-zerolag-input</div>
+      </div>
+    </div>
+  </div>
+);
+
+// ─── Title scene ───
+const TitleScene: React.FC = () => {
+  const frame = useCurrentFrame();
+  const { fps } = useVideoConfig();
+
+  const titleScale = spring({ frame, fps, config: { damping: 15, stiffness: 80 } });
+  const titleOpacity = interpolate(frame, [0, 20], [0, 1], { extrapolateRight: 'clamp' });
+  const subtitleOpacity = interpolate(frame, [15, 35], [0, 1], { extrapolateRight: 'clamp' });
+
+  return (
+    <AbsoluteFill style={{ background: colors.bg.dark, justifyContent: 'center', alignItems: 'center' }}>
+      <div style={{ textAlign: 'center', transform: `scale(${titleScale})` }}>
+        <div
+          style={{
+            fontSize: 80,
+            fontWeight: 700,
+            fontFamily: fonts.ui,
+            color: '#fff',
+            opacity: titleOpacity,
+            letterSpacing: -1.5,
+          }}
+        >
+          Zerolag Input
+        </div>
+        <div
+          style={{
+            fontSize: 30,
+            fontFamily: fonts.ui,
+            color: colors.text.dim,
+            opacity: subtitleOpacity,
+            marginTop: 16,
+          }}
+        >
+          Local echo for remote terminal sessions
+        </div>
+      </div>
+    </AbsoluteFill>
+  );
+};
+
+// ─── Outro ───
+const OutroScene: React.FC = () => {
+  const frame = useCurrentFrame();
+  const opacity = interpolate(frame, [0, 20], [0, 1], { extrapolateRight: 'clamp' });
+
+  return (
+    <AbsoluteFill style={{ background: colors.bg.dark, justifyContent: 'center', alignItems: 'center', opacity }}>
+      <div style={{ textAlign: 'center' }}>
+        <div style={{ fontSize: 56, fontWeight: 700, fontFamily: fonts.ui, color: '#fff' }}>Codeman</div>
+        <div style={{ fontSize: 26, fontFamily: fonts.ui, color: colors.accent.green, marginTop: 10 }}>
+          Zero-latency mobile input
+        </div>
+        <div style={{ fontSize: 16, fontFamily: fonts.mono, color: colors.text.muted, marginTop: 20 }}>
+          npm i xterm-zerolag-input
+        </div>
+      </div>
+    </AbsoluteFill>
+  );
+};
+
+// ─── Typing demo scene ───
+const TypingDemo: React.FC = () => {
+  const frame = useCurrentFrame();
+
+  const laggyTyped = computeVisibleText(frame, true);
+  const zerolagTyped = computeVisibleText(frame, false);
+
+  let activeKey: string | undefined;
+  let pressAge = 99;
+  for (let i = TYPING_SCHEDULE.length - 1; i >= 0; i--) {
+    const ev = TYPING_SCHEDULE[i];
+    if (frame >= ev.frame && frame < ev.frame + 5) {
+      activeKey = ev.char;
+      pressAge = frame - ev.frame;
+      break;
+    }
+  }
+
+  return (
+    <AbsoluteFill style={{ background: colors.bg.dark, justifyContent: 'center', alignItems: 'center' }}>
+      <div
+        style={{
+          display: 'flex',
+          gap: 50,
+          alignItems: 'flex-start',
+          transform: `scale(${PHONE_SCALE})`,
+          transformOrigin: 'center center',
+        }}
+      >
+        <div>
+          <PhoneLabel
+            title="With Zerolag"
+            detail="0ms local echo"
+            dotColor={colors.accent.green}
+            detailColor={colors.accent.green}
+          />
+          <MobileCodeman typed={zerolagTyped} cursorVisible activeKey={activeKey} pressAge={pressAge} noAnimation showPromo />
+        </div>
+        <div>
+          <PhoneLabel
+            title="Without Zerolag"
+            detail="600ms–2.7s server echo"
+            dotColor={colors.accent.red}
+            detailColor={colors.accent.red}
+          />
+          <MobileCodeman typed={laggyTyped} cursorVisible activeKey={activeKey} pressAge={pressAge} noAnimation />
+        </div>
+      </div>
+    </AbsoluteFill>
+  );
+};
+
+// ─── Phones entrance ───
+const PanelsEntrance: React.FC = () => {
+  const frame = useCurrentFrame();
+  const { fps } = useVideoConfig();
+
+  const scale = spring({ frame, fps, config: { damping: 15, stiffness: 80 } });
+  const opacity = interpolate(frame, [0, 10], [0, 1], { extrapolateRight: 'clamp' });
+
+  return (
+    <AbsoluteFill
+      style={{
+        background: colors.bg.dark,
+        justifyContent: 'center',
+        alignItems: 'center',
+        opacity,
+        transform: `scale(${scale * PHONE_SCALE})`,
+      }}
+    >
+      <div style={{ display: 'flex', gap: 50, alignItems: 'flex-start' }}>
+        <div>
+          <PhoneLabel
+            title="With Zerolag"
+            detail="0ms local echo"
+            dotColor={colors.accent.green}
+            detailColor={colors.accent.green}
+          />
+          <MobileCodeman typed="" cursorVisible noAnimation />
+        </div>
+        <div>
+          <PhoneLabel
+            title="Without Zerolag"
+            detail="600ms–2.7s server echo"
+            dotColor={colors.accent.red}
+            detailColor={colors.accent.red}
+          />
+          <MobileCodeman typed="" cursorVisible noAnimation />
+        </div>
+      </div>
+    </AbsoluteFill>
+  );
+};
+
+// ─── Main composition ───
+export const ZerolagDemo: React.FC = () => {
+  return (
+    <AbsoluteFill style={{ background: colors.bg.dark }}>
+      <Sequence from={TITLE_START} durationInFrames={TITLE_DUR}>
+        <TitleScene />
+      </Sequence>
+      <Sequence from={PHONES_START} durationInFrames={PHONES_DUR} premountFor={5}>
+        <PanelsEntrance />
+      </Sequence>
+      <Sequence from={TYPING_START} durationInFrames={TYPING_DUR + HOLD_DUR} premountFor={5}>
+        <TypingDemo />
+      </Sequence>
+      <Sequence from={OUTRO_START} durationInFrames={OUTRO_DUR} premountFor={5}>
+        <OutroScene />
+      </Sequence>
+    </AbsoluteFill>
+  );
+};

scripts/tunnel.sh

@@ -1,45 +1,209 @@
 #!/usr/bin/env bash
-# Quick Cloudflare Tunnel for Codeman
-# Usage: ./scripts/tunnel.sh [start|stop|status|url]
+# Cloudflare Tunnel manager for Codeman
+# Usage: ./scripts/tunnel.sh [quick|named] [start|stop|status|url]
+#
+# Modes:
+#   quick  — Quick tunnel with random trycloudflare.com URL (default)
+#   named  — Named tunnel on a fixed hostname (requires setup, see below)
+#
+# Environment variables:
+#   CLOUDFLARED_TUNNEL_NAME     — tunnel name (default: codeman)
+#   CLOUDFLARED_TUNNEL_ID       — tunnel UUID (from: cloudflared tunnel list)
+#   CODEMAN_TUNNEL_HOSTNAME     — public hostname (e.g. codeman.example.com)
+#
+# First-time named tunnel setup:
+#   cloudflared tunnel login
+#   cloudflared tunnel create <tunnel-name>
+#   cloudflared tunnel route dns <tunnel-name> <hostname>
+#   ./scripts/tunnel.sh named setup   # writes ~/.cloudflared/<tunnel-name>.yml
 set -euo pipefail
 
-SERVICE="codeman-tunnel"
+QUICK_SERVICE="codeman-tunnel"
+NAMED_SERVICE="codeman-tunnel-named"
+TUNNEL_NAME="${CLOUDFLARED_TUNNEL_NAME:-codeman}"
+TUNNEL_HOSTNAME="${CODEMAN_TUNNEL_HOSTNAME:-codeman.example.com}"
+CODEMAN_PORT="3000"
+LOG_FILE="$HOME/.codeman/tunnel.log"
 
-case "${1:-start}" in
-  start)
-    if ! systemctl --user is-active "$SERVICE" &>/dev/null; then
-      # Install service if not already
-      if ! systemctl --user cat "$SERVICE" &>/dev/null 2>&1; then
-        cp "$(dirname "$0")/codeman-tunnel.service" "$HOME/.config/systemd/user/"
-        systemctl --user daemon-reload
-      fi
-      systemctl --user start "$SERVICE"
-      echo "Tunnel starting... waiting for URL"
-      sleep 6
-    fi
-    # Extract the tunnel URL from journal
-    URL=$(grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' "$HOME/.codeman/tunnel.log" 2>/dev/null | tail -1)
-    if [ -n "$URL" ]; then
-      echo "$URL"
-    else
-      echo "URL not ready yet, try: $0 url"
-    fi
+# ── helpers ──────────────────────────────────────────────────────────────────
+
+_require_cloudflared() {
+  if ! command -v cloudflared &>/dev/null; then
+    echo "Error: cloudflared not found. Install with: yay -S cloudflared" >&2
+    exit 1
+  fi
+}
+
+_cloudflared_bin() {
+  command -v cloudflared
+}
+
+_install_service() {
+  local svc_file="$1"
+  local svc_name="$2"
+  if ! systemctl --user cat "$svc_name" &>/dev/null 2>&1; then
+    cp "$(dirname "$0")/$svc_file" "$HOME/.config/systemd/user/"
+    systemctl --user daemon-reload
+    echo "Service $svc_name installed."
+  fi
+}
+
+_install_named_service() {
+  if ! systemctl --user cat "$NAMED_SERVICE" &>/dev/null 2>&1; then
+    # Generate service file with the configured tunnel name
+    sed "s/codeman\.yml/$TUNNEL_NAME.yml/g; s/run codeman/run $TUNNEL_NAME/g" \
+      "$(dirname "$0")/codeman-tunnel-named.service" \
+      > "$HOME/.config/systemd/user/codeman-tunnel-named.service"
+    systemctl --user daemon-reload
+    echo "Service $NAMED_SERVICE installed (tunnel: $TUNNEL_NAME)."
+  fi
+}
+
+# ── named tunnel setup ───────────────────────────────────────────────────────
+
+_named_setup() {
+  _require_cloudflared
+
+  local creds_dir="$HOME/.cloudflared"
+  local config_file="$creds_dir/$TUNNEL_NAME.yml"
+  # Replace with your tunnel ID (from: cloudflared tunnel list)
+  local tunnel_id="${CLOUDFLARED_TUNNEL_ID:-YOUR_TUNNEL_ID_HERE}"
+  local creds_file="$creds_dir/$tunnel_id.json"
+
+  if [ ! -f "$creds_file" ]; then
+    echo "Credentials not found: $creds_file"
+    echo "Run: cloudflared tunnel create $TUNNEL_NAME"
+    exit 1
+  fi
+
+  cat > "$config_file" <<EOF
+tunnel: $tunnel_id
+credentials-file: $creds_file
+
+ingress:
+  - hostname: $TUNNEL_HOSTNAME
+    service: http://localhost:$CODEMAN_PORT
+  - service: http_status:404
+EOF
+
+  echo "Config written to $config_file"
+  echo "Tunnel ID: $tunnel_id"
+  echo "Hostname:  $TUNNEL_HOSTNAME"
+  echo ""
+  echo "Next steps:"
+  echo "  1. Add Cloudflare Access policy for $TUNNEL_HOSTNAME (Zero Trust dashboard)"
+  echo "  2. ./scripts/tunnel.sh named start"
+}
+
+# ── quick mode ───────────────────────────────────────────────────────────────
+
+_quick_start() {
+  if ! systemctl --user is-active "$QUICK_SERVICE" &>/dev/null; then
+    _install_service "codeman-tunnel.service" "$QUICK_SERVICE"
+    systemctl --user start "$QUICK_SERVICE"
+    echo "Quick tunnel starting... waiting for URL"
+    sleep 6
+  fi
+  local url
+  url=$(grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' "$LOG_FILE" 2>/dev/null | tail -1)
+  if [ -n "$url" ]; then
+    echo "$url"
+  else
+    echo "URL not ready yet, try: $0 quick url"
+  fi
+}
+
+_quick_stop() {
+  systemctl --user stop "$QUICK_SERVICE"
+  echo "Quick tunnel stopped"
+}
+
+_quick_status() {
+  systemctl --user status "$QUICK_SERVICE" --no-pager 2>&1 | head -10
+  echo ""
+  echo "URL:"
+  grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' "$LOG_FILE" 2>/dev/null | tail -1
+}
+
+_quick_url() {
+  grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' "$LOG_FILE" 2>/dev/null | tail -1
+}
+
+# ── named mode ───────────────────────────────────────────────────────────────
+
+_named_start() {
+  _require_cloudflared
+  if [ ! -f "$HOME/.cloudflared/$TUNNEL_NAME.yml" ]; then
+    echo "Config not found. Run: $0 named setup"
+    exit 1
+  fi
+  if ! systemctl --user is-active "$NAMED_SERVICE" &>/dev/null; then
+    _install_named_service
+    systemctl --user start "$NAMED_SERVICE"
+    echo "Named tunnel starting..."
+    sleep 3
+  fi
+  echo "https://$TUNNEL_HOSTNAME"
+}
+
+_named_stop() {
+  systemctl --user stop "$NAMED_SERVICE"
+  echo "Named tunnel stopped"
+}
+
+_named_status() {
+  systemctl --user status "$NAMED_SERVICE" --no-pager 2>&1 | head -10
+  echo ""
+  echo "URL: https://$TUNNEL_HOSTNAME"
+}
+
+_named_enable() {
+  _install_named_service
+  systemctl --user enable "$NAMED_SERVICE"
+  echo "Named tunnel enabled at boot."
+}
+
+_named_disable() {
+  systemctl --user disable "$NAMED_SERVICE"
+  echo "Named tunnel disabled."
+}
+
+# ── dispatch ─────────────────────────────────────────────────────────────────
+
+MODE="${1:-quick}"
+CMD="${2:-start}"
+
+case "$MODE" in
+  quick)
+    case "$CMD" in
+      start)  _quick_start ;;
+      stop)   _quick_stop ;;
+      status) _quick_status ;;
+      url)    _quick_url ;;
+      *) echo "Usage: $0 quick [start|stop|status|url]"; exit 1 ;;
+    esac
     ;;
-  stop)
-    systemctl --user stop "$SERVICE"
-    echo "Tunnel stopped"
-    ;;
-  status)
-    systemctl --user status "$SERVICE" --no-pager 2>&1 | head -10
-    echo ""
-    echo "URL:"
-    grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' "$HOME/.codeman/tunnel.log" 2>/dev/null | tail -1
-    ;;
-  url)
-    grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' "$HOME/.codeman/tunnel.log" 2>/dev/null | tail -1
+  named)
+    case "$CMD" in
+      start)   _named_start ;;
+      stop)    _named_stop ;;
+      status)  _named_status ;;
+      url)     echo "https://$TUNNEL_HOSTNAME" ;;
+      setup)   _named_setup ;;
+      enable)  _named_enable ;;
+      disable) _named_disable ;;
+      *) echo "Usage: $0 named [start|stop|status|url|setup|enable|disable]"; exit 1 ;;
+    esac
     ;;
+  # backward compat: no mode prefix → quick tunnel
+  start)  _quick_start ;;
+  stop)   _quick_stop ;;
+  status) _quick_status ;;
+  url)    _quick_url ;;
   *)
-    echo "Usage: $0 [start|stop|status|url]"
+    echo "Usage: $0 [quick|named] [start|stop|status|url]"
+    echo "       $0 named setup    # first-time named tunnel configuration"
+    echo "       $0 named enable   # start at boot"
     exit 1
     ;;
 esac

src/ai-checker-base.ts

@@ -28,8 +28,9 @@ import { existsSync, readFileSync, unlinkSync, writeFileSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 import { EventEmitter } from 'node:events';
-import { getAugmentedPath } from './utils/claude-cli-resolver.js';
-import { ANSI_ESCAPE_PATTERN_SIMPLE } from './utils/index.js';
+import { getAugmentedPath, ANSI_ESCAPE_PATTERN_SIMPLE } from './utils/index.js';
+import { AI_CHECK_MAX_BACKOFF_MS } from './config/ai-defaults.js';
+import { getErrorMessage } from './types.js';
 
 // ========== Security Validation ==========
 
@@ -293,7 +294,7 @@ export abstract class AiCheckerBase<
       this.emit('checkCompleted', result);
       return result;
     } catch (err) {
-      const errorMsg = err instanceof Error ? err.message : String(err);
+      const errorMsg = getErrorMessage(err);
       this.handleError(errorMsg);
       const result = this.createErrorResult(errorMsg, Date.now() - this.checkStartTime);
       this.emit('checkFailed', errorMsg);
@@ -412,9 +413,7 @@ export abstract class AiCheckerBase<
       });
       muxProcess.unref();
     } catch (err) {
-      throw new Error(
-        `Failed to spawn ${this.checkDescription} tmux session: ${err instanceof Error ? err.message : String(err)}`
-      );
+      throw new Error(`Failed to spawn ${this.checkDescription} tmux session: ${getErrorMessage(err)}`);
     }
 
     // Poll the temp file for completion
@@ -534,10 +533,7 @@ export abstract class AiCheckerBase<
       // P1-005: Exponential backoff for errors
       // Base cooldown * 2^(consecutiveErrors-1), capped at 5 minutes
       const backoffMultiplier = Math.pow(2, this.consecutiveErrors - 1);
-      const backoffCooldownMs = Math.min(
-        this.config.errorCooldownMs * backoffMultiplier,
-        5 * 60 * 1000 // Max 5 minutes
-      );
+      const backoffCooldownMs = Math.min(this.config.errorCooldownMs * backoffMultiplier, AI_CHECK_MAX_BACKOFF_MS);
       this.log(`Exponential backoff: ${Math.round(backoffCooldownMs / 1000)}s (error #${this.consecutiveErrors})`);
       this.startCooldown(backoffCooldownMs);
     }

src/ai-idle-checker.ts

@@ -30,6 +30,14 @@ import {
   type AiCheckerResultBase,
   type AiCheckerStateBase,
 } from './ai-checker-base.js';
+import {
+  AI_CHECK_MODEL,
+  AI_IDLE_CHECK_MAX_CONTEXT,
+  AI_IDLE_CHECK_TIMEOUT_MS,
+  AI_IDLE_CHECK_COOLDOWN_MS,
+  AI_IDLE_CHECK_ERROR_COOLDOWN_MS,
+  AI_CHECK_MAX_CONSECUTIVE_ERRORS,
+} from './config/ai-defaults.js';
 
 // ========== Types ==========
 
@@ -45,12 +53,12 @@ export type AiCheckState = AiCheckerStateBase<AiCheckVerdict>;
 
 const DEFAULT_AI_CHECK_CONFIG: AiIdleCheckConfig = {
   enabled: true,
-  model: 'claude-opus-4-5-20251101',
-  maxContextChars: 16000,
-  checkTimeoutMs: 90000,
-  cooldownMs: 180000,
-  errorCooldownMs: 60000,
-  maxConsecutiveErrors: 3,
+  model: AI_CHECK_MODEL,
+  maxContextChars: AI_IDLE_CHECK_MAX_CONTEXT,
+  checkTimeoutMs: AI_IDLE_CHECK_TIMEOUT_MS,
+  cooldownMs: AI_IDLE_CHECK_COOLDOWN_MS,
+  errorCooldownMs: AI_IDLE_CHECK_ERROR_COOLDOWN_MS,
+  maxConsecutiveErrors: AI_CHECK_MAX_CONSECUTIVE_ERRORS,
 };
 
 /** Pattern to match IDLE or WORKING as the first word of output */

src/ai-plan-checker.ts

@@ -29,6 +29,14 @@ import {
   type AiCheckerResultBase,
   type AiCheckerStateBase,
 } from './ai-checker-base.js';
+import {
+  AI_CHECK_MODEL,
+  AI_PLAN_CHECK_MAX_CONTEXT,
+  AI_PLAN_CHECK_TIMEOUT_MS,
+  AI_PLAN_CHECK_COOLDOWN_MS,
+  AI_PLAN_CHECK_ERROR_COOLDOWN_MS,
+  AI_CHECK_MAX_CONSECUTIVE_ERRORS,
+} from './config/ai-defaults.js';
 
 // ========== Types ==========
 
@@ -44,12 +52,12 @@ export type AiPlanCheckState = AiCheckerStateBase<AiPlanCheckVerdict>;
 
 const DEFAULT_PLAN_CHECK_CONFIG: AiPlanCheckConfig = {
   enabled: true,
-  model: 'claude-opus-4-5-20251101',
-  maxContextChars: 8000,
-  checkTimeoutMs: 60000,
-  cooldownMs: 30000,
-  errorCooldownMs: 30000,
-  maxConsecutiveErrors: 3,
+  model: AI_CHECK_MODEL,
+  maxContextChars: AI_PLAN_CHECK_MAX_CONTEXT,
+  checkTimeoutMs: AI_PLAN_CHECK_TIMEOUT_MS,
+  cooldownMs: AI_PLAN_CHECK_COOLDOWN_MS,
+  errorCooldownMs: AI_PLAN_CHECK_ERROR_COOLDOWN_MS,
+  maxConsecutiveErrors: AI_CHECK_MAX_CONSECUTIVE_ERRORS,
 };
 
 /** Pattern to match PLAN_MODE or NOT_PLAN_MODE as the first word(s) of output */

src/bash-tool-parser.ts

@@ -15,6 +15,7 @@
 import { EventEmitter } from 'node:events';
 import { v4 as uuidv4 } from 'uuid';
 import { ActiveBashTool } from './types.js';
+import { CleanupManager, Debouncer, stripAnsi } from './utils/index.js';
 
 // ========== Configuration Constants ==========
 
@@ -145,15 +146,14 @@ export class BashToolParser extends EventEmitter<BashToolParserEvents> {
   private _workingDir: string;
   private _homeDir: string;
 
-  // Track auto-remove timers for cleanup
-  private _autoRemoveTimers: Set<ReturnType<typeof setTimeout>> = new Set();
+  // Centralized resource cleanup for auto-remove timers
+  private cleanup = new CleanupManager();
 
   // Flag to prevent operations after destroy
   private _destroyed: boolean = false;
 
   // Debouncing
-  private _pendingUpdate: boolean = false;
-  private _updateTimer: ReturnType<typeof setTimeout> | null = null;
+  private _updateDeb = new Debouncer(EVENT_DEBOUNCE_MS);
 
   constructor(config: BashToolParserConfig) {
     super();
@@ -462,7 +462,7 @@ export class BashToolParser extends EventEmitter<BashToolParserEvents> {
    * Process a single line of terminal output (raw — will strip ANSI).
    */
   private processLine(line: string): void {
-    const cleanLine = this.stripAnsi(line);
+    const cleanLine = stripAnsi(line);
     this.processCleanLine(cleanLine);
   }
 
@@ -470,109 +470,91 @@ export class BashToolParser extends EventEmitter<BashToolParserEvents> {
    * Process a single pre-stripped line of terminal output.
    */
   private processCleanLine(cleanLine: string): void {
-    // Check for tool start
+    if (this._handleToolStart(cleanLine)) return;
+    if (this._handleToolCompletion(cleanLine)) return;
+    if (this._handleTextCommand(cleanLine)) return;
+    this._handleLogFileMention(cleanLine);
+  }
+
+  private _handleToolStart(cleanLine: string): boolean {
     const startMatch = cleanLine.match(BASH_TOOL_START_PATTERN);
-    if (startMatch) {
-      const command = startMatch[1];
-      const timeout = startMatch[2]?.trim();
+    if (!startMatch) return false;
 
-      // Check if this is a file-viewing command
-      if (this.isFileViewerCommand(command)) {
-        const filePaths = this.extractFilePaths(command);
+    const command = startMatch[1];
+    const timeout = startMatch[2]?.trim();
 
-        // Skip if any file path is already tracked (cross-pattern dedup)
-        if (filePaths.some((fp) => this.isFilePathTracked(fp))) {
-          return;
-        }
+    if (!this.isFileViewerCommand(command)) return true;
 
-        if (filePaths.length > 0) {
-          const tool: ActiveBashTool = {
-            id: uuidv4(),
-            command,
-            filePaths,
-            timeout,
-            startedAt: Date.now(),
-            status: 'running',
-            sessionId: this._sessionId,
-          };
+    const filePaths = this.extractFilePaths(command);
 
-          // Enforce max tools limit
-          if (this._activeTools.size >= MAX_ACTIVE_TOOLS) {
-            // Remove oldest tool
-            const oldest = Array.from(this._activeTools.entries()).sort((a, b) => a[1].startedAt - b[1].startedAt)[0];
-            if (oldest) {
-              this._activeTools.delete(oldest[0]);
-            }
+    // Skip if any file path is already tracked (cross-pattern dedup)
+    if (filePaths.some((fp) => this.isFilePathTracked(fp))) return true;
+
+    if (filePaths.length > 0) {
+      const tool = this._createActiveTool(command, filePaths, 'running', timeout);
+
+      // Enforce max tools limit
+      if (this._activeTools.size >= MAX_ACTIVE_TOOLS) {
+        // Remove oldest tool (O(n) min-scan instead of O(n log n) sort)
+        let oldestKey: string | undefined;
+        let oldestTime = Infinity;
+        for (const [key, entry] of this._activeTools) {
+          if (entry.startedAt < oldestTime) {
+            oldestTime = entry.startedAt;
+            oldestKey = key;
           }
-
-          this._activeTools.set(tool.id, tool);
-          this._lastToolId = tool.id;
-
-          this.emit('toolStart', tool);
-          this.scheduleUpdate();
         }
-      }
-      return;
-    }
-
-    // Check for tool completion
-    if (TOOL_COMPLETION_PATTERN.test(cleanLine) && this._lastToolId) {
-      const tool = this._activeTools.get(this._lastToolId);
-      if (tool && tool.status === 'running') {
-        tool.status = 'completed';
-        this.emit('toolEnd', tool);
-        this.scheduleUpdate();
-
-        // Remove completed tool after a short delay to allow UI to show completion
-        const timer = setTimeout(() => {
-          this._autoRemoveTimers.delete(timer);
-          if (this._destroyed) return;
-          this._activeTools.delete(tool.id);
-          this.scheduleUpdate();
-        }, 2000);
-        this._autoRemoveTimers.add(timer);
-      }
-      this._lastToolId = null;
-      return;
-    }
-
-    // Fallback: Check for command suggestions in plain text (e.g., "tail -f /tmp/file.log")
-    const textCmdMatch = cleanLine.match(TEXT_COMMAND_PATTERN);
-    if (textCmdMatch) {
-      const filePath = textCmdMatch[2];
-
-      // Create a suggestion tool (marked as 'suggestion' status)
-      const tool: ActiveBashTool = {
-        id: uuidv4(),
-        command: cleanLine.trim(),
-        filePaths: [filePath],
-        timeout: undefined,
-        startedAt: Date.now(),
-        status: 'running', // Shows as clickable
-        sessionId: this._sessionId,
-      };
-
-      // Don't add if file path already tracked (cross-pattern dedup)
-      if (this.isFilePathTracked(filePath)) {
-        return;
+        if (oldestKey) {
+          this._activeTools.delete(oldestKey);
+        }
       }
 
       this._activeTools.set(tool.id, tool);
+      this._lastToolId = tool.id;
+
       this.emit('toolStart', tool);
       this.scheduleUpdate();
-
-      // Auto-remove suggestions after 30 seconds
-      const timer = setTimeout(() => {
-        this._autoRemoveTimers.delete(timer);
-        if (this._destroyed) return;
-        this._activeTools.delete(tool.id);
-        this.scheduleUpdate();
-      }, 30000);
-      this._autoRemoveTimers.add(timer);
-      return;
     }
 
-    // Last fallback: Check for log file paths mentioned anywhere in the line
+    return true;
+  }
+
+  private _handleToolCompletion(cleanLine: string): boolean {
+    if (!TOOL_COMPLETION_PATTERN.test(cleanLine) || !this._lastToolId) return false;
+
+    const tool = this._activeTools.get(this._lastToolId);
+    if (tool && tool.status === 'running') {
+      tool.status = 'completed';
+      this.emit('toolEnd', tool);
+      this.scheduleUpdate();
+
+      this._scheduleAutoRemove(tool.id, 2000, 'auto-remove completed tool');
+    }
+    this._lastToolId = null;
+    return true;
+  }
+
+  private _handleTextCommand(cleanLine: string): boolean {
+    const textCmdMatch = cleanLine.match(TEXT_COMMAND_PATTERN);
+    if (!textCmdMatch) return false;
+
+    const filePath = textCmdMatch[2];
+
+    // Don't add if file path already tracked (cross-pattern dedup)
+    if (this.isFilePathTracked(filePath)) return true;
+
+    const tool = this._createActiveTool(cleanLine.trim(), [filePath], 'running');
+
+    this._activeTools.set(tool.id, tool);
+    this.emit('toolStart', tool);
+    this.scheduleUpdate();
+
+    // Auto-remove suggestions after 30 seconds
+    this._scheduleAutoRemove(tool.id, 30000, 'auto-remove suggestion tool');
+    return true;
+  }
+
+  private _handleLogFileMention(cleanLine: string): void {
     LOG_FILE_MENTION_PATTERN.lastIndex = 0;
     let logMatch;
     while ((logMatch = LOG_FILE_MENTION_PATTERN.exec(cleanLine)) !== null) {
@@ -584,31 +566,46 @@ export class BashToolParser extends EventEmitter<BashToolParserEvents> {
       // Skip if file path already tracked (cross-pattern dedup)
       if (this.isFilePathTracked(filePath)) continue;
 
-      const tool: ActiveBashTool = {
-        id: uuidv4(),
-        command: `View: ${filePath}`,
-        filePaths: [filePath],
-        timeout: undefined,
-        startedAt: Date.now(),
-        status: 'running',
-        sessionId: this._sessionId,
-      };
+      const tool = this._createActiveTool(`View: ${filePath}`, [filePath], 'running');
 
       this._activeTools.set(tool.id, tool);
       this.emit('toolStart', tool);
       this.scheduleUpdate();
 
       // Auto-remove after 60 seconds
-      const timer = setTimeout(() => {
-        this._autoRemoveTimers.delete(timer);
-        if (this._destroyed) return;
-        this._activeTools.delete(tool.id);
-        this.scheduleUpdate();
-      }, 60000);
-      this._autoRemoveTimers.add(timer);
+      this._scheduleAutoRemove(tool.id, 60000, 'auto-remove log file tool');
     }
   }
 
+  private _createActiveTool(
+    command: string,
+    filePaths: string[],
+    status: ActiveBashTool['status'],
+    timeout?: string
+  ): ActiveBashTool {
+    return {
+      id: uuidv4(),
+      command,
+      filePaths,
+      timeout,
+      startedAt: Date.now(),
+      status,
+      sessionId: this._sessionId,
+    };
+  }
+
+  private _scheduleAutoRemove(toolId: string, delayMs: number, description: string): void {
+    this.cleanup.setTimeout(
+      () => {
+        if (this._destroyed) return;
+        this._activeTools.delete(toolId);
+        this.scheduleUpdate();
+      },
+      delayMs,
+      { description }
+    );
+  }
+
   /**
    * Check if a command is a file-viewing command worth tracking.
    */
@@ -662,26 +659,13 @@ export class BashToolParser extends EventEmitter<BashToolParserEvents> {
     return this.deduplicatePaths(rawPaths);
   }
 
-  /**
-   * Strip ANSI escape codes from a string.
-   */
-  private stripAnsi(str: string): string {
-    // Comprehensive ANSI pattern
-    // eslint-disable-next-line no-control-regex
-    return str.replace(/\x1b(?:\[[0-9;?]*[A-Za-z]|\][^\x07\x1b]*(?:\x07|\x1b\\)|[=>])/g, '');
-  }
-
   /**
    * Schedule a debounced update emission.
    */
   private scheduleUpdate(): void {
-    if (this._pendingUpdate) return;
-
-    this._pendingUpdate = true;
-    this._updateTimer = setTimeout(() => {
-      this._pendingUpdate = false;
+    this._updateDeb.schedule(() => {
       this.emitUpdate();
-    }, EVENT_DEBOUNCE_MS);
+    });
   }
 
   /**
@@ -696,15 +680,8 @@ export class BashToolParser extends EventEmitter<BashToolParserEvents> {
    */
   destroy(): void {
     this._destroyed = true;
-    if (this._updateTimer) {
-      clearTimeout(this._updateTimer);
-      this._updateTimer = null;
-    }
-    // Clear all auto-remove timers to prevent orphaned callbacks
-    for (const timer of this._autoRemoveTimers) {
-      clearTimeout(timer);
-    }
-    this._autoRemoveTimers.clear();
+    this._updateDeb.dispose();
+    this.cleanup.dispose();
     this._activeTools.clear();
     this.removeAllListeners();
   }

src/config/ai-defaults.ts

@@ -0,0 +1,58 @@
+/**
+ * @fileoverview Default model, context limits, and timing for AI-powered checkers.
+ *
+ * Centralizes the AI model identifier, context window sizes, and timeout/cooldown
+ * defaults used by the idle checker, plan checker, respawn controller defaults,
+ * and respawn route fallbacks. Change values here when tuning AI check behavior.
+ *
+ * @module config/ai-defaults
+ */
+
+// ============================================================================
+// Model & Context
+// ============================================================================
+
+/** Default model for AI idle and plan checkers */
+export const AI_CHECK_MODEL = 'claude-opus-4-5-20251101';
+
+/** Max context chars for idle checker (~4k tokens) */
+export const AI_IDLE_CHECK_MAX_CONTEXT = 16000;
+
+/** Max context chars for plan checker (~2k tokens, plan mode UI is compact) */
+export const AI_PLAN_CHECK_MAX_CONTEXT = 8000;
+
+// ============================================================================
+// AI Idle Checker Timing
+// ============================================================================
+
+/** Timeout for AI idle check (90 seconds — thinking can be slow) */
+export const AI_IDLE_CHECK_TIMEOUT_MS = 90_000;
+
+/** Cooldown after WORKING verdict (3 minutes) */
+export const AI_IDLE_CHECK_COOLDOWN_MS = 180_000;
+
+/** Cooldown after AI idle check error (1 minute) */
+export const AI_IDLE_CHECK_ERROR_COOLDOWN_MS = 60_000;
+
+// ============================================================================
+// AI Plan Checker Timing
+// ============================================================================
+
+/** Timeout for AI plan check (60 seconds — allows time for thinking) */
+export const AI_PLAN_CHECK_TIMEOUT_MS = 60_000;
+
+/** Cooldown after NOT_PLAN_MODE verdict (30 seconds) */
+export const AI_PLAN_CHECK_COOLDOWN_MS = 30_000;
+
+/** Cooldown after AI plan check error (30 seconds) */
+export const AI_PLAN_CHECK_ERROR_COOLDOWN_MS = 30_000;
+
+// ============================================================================
+// Shared AI Checker Limits
+// ============================================================================
+
+/** Max consecutive errors before disabling an AI checker */
+export const AI_CHECK_MAX_CONSECUTIVE_ERRORS = 3;
+
+/** Maximum exponential backoff cap for AI checker errors (5 minutes) */
+export const AI_CHECK_MAX_BACKOFF_MS = 5 * 60 * 1000;

src/config/auth-config.ts

@@ -0,0 +1,35 @@
+/**
+ * @fileoverview Authentication, rate limiting, and hook security constants.
+ *
+ * Controls auth session lifecycle, brute-force protection,
+ * and Claude Code hook timeouts.
+ *
+ * @module config/auth-config
+ */
+
+// ============================================================================
+// Session Cookies
+// ============================================================================
+
+/** Auth session cookie TTL — matches autonomous run length (ms) */
+export const AUTH_SESSION_TTL_MS = 24 * 60 * 60 * 1000;
+
+/** Max concurrent auth sessions per server */
+export const MAX_AUTH_SESSIONS = 100;
+
+// ============================================================================
+// Rate Limiting
+// ============================================================================
+
+/** Max failed auth attempts per IP before 429 rejection */
+export const AUTH_FAILURE_MAX = 10;
+
+/** Failed auth attempt tracking window (ms) */
+export const AUTH_FAILURE_WINDOW_MS = 15 * 60 * 1000;
+
+// ============================================================================
+// Hooks
+// ============================================================================
+
+/** Timeout for Claude Code hook curl commands (ms) */
+export const HOOK_TIMEOUT_MS = 10000;

src/config/buffer-limits.ts

@@ -22,14 +22,16 @@
  * Maximum terminal buffer size in characters.
  * Contains raw terminal output with ANSI escape sequences.
  * Reduced from 5MB to 2MB for better render performance.
+ * Override: CODEMAN_MAX_TERMINAL_BUFFER (bytes)
  */
-export const MAX_TERMINAL_BUFFER_SIZE = 2 * 1024 * 1024; // 2MB
+export const MAX_TERMINAL_BUFFER_SIZE = parseInt(process.env.CODEMAN_MAX_TERMINAL_BUFFER || '') || 2 * 1024 * 1024;
 
 /**
  * Size to trim terminal buffer to when max is exceeded.
  * Keeps the most recent portion to preserve context.
+ * Override: CODEMAN_TRIM_TERMINAL_TO (bytes)
  */
-export const TRIM_TERMINAL_TO = 1.5 * 1024 * 1024; // 1.5MB
+export const TRIM_TERMINAL_TO = parseInt(process.env.CODEMAN_TRIM_TERMINAL_TO || '') || 1.5 * 1024 * 1024;
 
 // ============================================================================
 // Text Output Buffer Limits
@@ -38,13 +40,15 @@ export const TRIM_TERMINAL_TO = 1.5 * 1024 * 1024; // 1.5MB
 /**
  * Maximum text output buffer size in characters.
  * Contains ANSI-stripped text for search and analysis.
+ * Override: CODEMAN_MAX_TEXT_OUTPUT (bytes)
  */
-export const MAX_TEXT_OUTPUT_SIZE = 1 * 1024 * 1024; // 1MB
+export const MAX_TEXT_OUTPUT_SIZE = parseInt(process.env.CODEMAN_MAX_TEXT_OUTPUT || '') || 1 * 1024 * 1024;
 
 /**
  * Size to trim text output buffer to when max is exceeded.
+ * Override: CODEMAN_TRIM_TEXT_TO (bytes)
  */
-export const TRIM_TEXT_TO = 768 * 1024; // 768KB
+export const TRIM_TEXT_TO = parseInt(process.env.CODEMAN_TRIM_TEXT_TO || '') || 768 * 1024;
 
 // ============================================================================
 // Message Buffer Limits
@@ -53,13 +57,9 @@ export const TRIM_TEXT_TO = 768 * 1024; // 768KB
 /**
  * Maximum number of Claude JSON messages to keep in memory per session.
  * Older messages are discarded when limit is exceeded.
+ * Override: CODEMAN_MAX_MESSAGES (count)
  */
-export const MAX_MESSAGES = 1000;
-
-/**
- * Number of messages to keep when trimming (80% of max).
- */
-export const TRIM_MESSAGES_TO = 800;
+export const MAX_MESSAGES = parseInt(process.env.CODEMAN_MAX_MESSAGES || '') || 1000;
 
 // ============================================================================
 // Line Buffer Limits
@@ -85,3 +85,14 @@ export const MAX_RESPAWN_BUFFER_SIZE = 1 * 1024 * 1024; // 1MB
  * Size to trim respawn buffer to when max is exceeded.
  */
 export const TRIM_RESPAWN_BUFFER_TO = 512 * 1024; // 512KB
+
+// ============================================================================
+// File Peek Limits
+// ============================================================================
+
+/**
+ * Maximum bytes to read when peeking at the beginning of a file.
+ * Used with `createReadStream({ end })` (inclusive) to read the first 8KB,
+ * which is enough to extract metadata from the first few JSONL lines.
+ */
+export const FILE_PEEK_BYTES = 8 * 1024 - 1; // 8KB (inclusive end offset)

src/config/exec-timeout.ts

@@ -0,0 +1,10 @@
+/**
+ * @fileoverview Shared exec timeout constant.
+ *
+ * Used by CLI resolvers and tmux-manager for execSync/exec calls.
+ *
+ * @module config/exec-timeout
+ */
+
+/** Timeout for exec commands (5 seconds) */
+export const EXEC_TIMEOUT_MS = 5000;

src/config/map-limits.ts

@@ -47,6 +47,15 @@ export const MAX_TODOS_PER_SESSION = 500;
 // Pending Tool Calls Limits
 // ============================================================================
 
+// ============================================================================
+// Agent Tracking Limits
+// ============================================================================
+
+/**
+ * Maximum agents to track across all sessions (LRU eviction when exceeded).
+ */
+export const MAX_TRACKED_AGENTS = 500;
+
 /**
  * Maximum pending tool calls to track per subagent.
  * Entries should be cleaned up on tool_result, but this prevents leaks.

src/config/server-timing.ts

@@ -0,0 +1,88 @@
+/**
+ * @fileoverview Web server performance and scheduling constants.
+ *
+ * Controls terminal batching throughput, SSE health checking,
+ * state persistence debouncing, and scheduled run timing.
+ *
+ * @module config/server-timing
+ */
+
+// ============================================================================
+// Terminal & SSE Performance
+// ============================================================================
+
+/** Terminal data batching interval — targets 60fps (ms) */
+export const TERMINAL_BATCH_INTERVAL = 16;
+
+/** Immediate flush threshold for terminal batches (bytes).
+ * Set high (32KB) to allow effective batching; avg Ink events are ~14KB. */
+export const BATCH_FLUSH_THRESHOLD = 32 * 1024;
+
+/** Task event batching interval (ms) */
+export const TASK_UPDATE_BATCH_INTERVAL = 100;
+
+/** SSE heartbeat interval — sends padded keepalive to flush proxy buffers (ms).
+ * 15s is fast enough to keep Cloudflare tunnel buffers flushed while avoiding
+ * excessive bandwidth. Also serves as dead-client detection. */
+export const SSE_HEARTBEAT_INTERVAL = 15 * 1000;
+
+/** SSE padding size (bytes). Cloudflare quick tunnels buffer small SSE events;
+ * appending ~8KB of SSE comment padding forces the proxy to flush immediately.
+ * SSE comments (lines starting with ':') are silently ignored by EventSource. */
+export const SSE_PADDING_SIZE = 8 * 1024;
+
+// ============================================================================
+// State Persistence
+// ============================================================================
+
+/** State update debounce — batches expensive toDetailedState() calls (ms) */
+export const STATE_UPDATE_DEBOUNCE_INTERVAL = 500;
+
+/** Sessions list cache TTL — avoids re-serializing on every SSE init (ms) */
+export const SESSIONS_LIST_CACHE_TTL = 1000;
+
+// ============================================================================
+// Scheduled Runs
+// ============================================================================
+
+/** Scheduled runs cleanup check interval (ms) */
+export const SCHEDULED_CLEANUP_INTERVAL = 5 * 60 * 1000;
+
+/** Completed scheduled run max age before cleanup (ms) */
+export const SCHEDULED_RUN_MAX_AGE = 60 * 60 * 1000;
+
+/** Session limit retry wait before retrying (ms) */
+export const SESSION_LIMIT_WAIT_MS = 5000;
+
+/** Pause between scheduled run iterations (ms) */
+export const ITERATION_PAUSE_MS = 2000;
+
+// ============================================================================
+// Mux Stats
+// ============================================================================
+
+/** Mux stats collection interval (ms) */
+export const STATS_COLLECTION_INTERVAL_MS = 2000;
+
+// ============================================================================
+// Process Error Recovery
+// ============================================================================
+
+/** Max consecutive unhandled errors before auto-restart */
+export const MAX_CONSECUTIVE_ERRORS = 5;
+
+/** Error counter reset interval — forgives errors after quiet period (ms) */
+export const ERROR_RESET_MS = 60_000;
+
+// ============================================================================
+// Common Cleanup Intervals
+// ============================================================================
+
+/** Standard 1-minute cleanup/check interval used by multiple subsystems (ms) */
+export const CLEANUP_CHECK_INTERVAL_MS = 60_000;
+
+/** Standard 1-hour max age for stale/completed data (ms) */
+export const STALE_DATA_MAX_AGE_MS = 60 * 60 * 1000;
+
+/** Standard 5-minute inactivity timeout for streams and caches (ms) */
+export const INACTIVITY_TIMEOUT_MS = 5 * 60 * 1000;

src/config/team-config.ts

@@ -0,0 +1,18 @@
+/**
+ * @fileoverview Agent Teams polling and cache configuration.
+ *
+ * Controls how frequently TeamWatcher polls ~/.claude/teams/
+ * and how many teams/tasks are cached in memory.
+ *
+ * @module config/team-config
+ */
+
+/** Team directory poll interval (ms) */
+export const TEAM_POLL_INTERVAL_MS = 30_000;
+
+/** Max cached team configs (LRU eviction) */
+export const MAX_CACHED_TEAMS = 50;
+
+/** Max cached team tasks and inbox messages (LRU eviction).
+ * Used for both teamTasks and inboxCache maps. */
+export const MAX_CACHED_TASKS = 200;

src/config/terminal-limits.ts

@@ -0,0 +1,15 @@
+/**
+ * @fileoverview Terminal dimension and input validation limits.
+ *
+ * Used by API routes to validate resize, input, and session
+ * creation requests. Separate from buffer-limits.ts which
+ * controls memory buffer sizes.
+ *
+ * @module config/terminal-limits
+ */
+
+/** Max input length per API request (bytes) */
+export const MAX_INPUT_LENGTH = 64 * 1024;
+
+/** Max session name length (chars) */
+export const MAX_SESSION_NAME_LENGTH = 128;

src/config/tunnel-config.ts

@@ -0,0 +1,47 @@
+/**
+ * @fileoverview Cloudflare tunnel and QR authentication constants.
+ *
+ * Controls QR token rotation timing, rate limiting,
+ * and tunnel process lifecycle.
+ *
+ * @module config/tunnel-config
+ */
+
+// ============================================================================
+// QR Token Rotation
+// ============================================================================
+
+/** QR token auto-rotation interval (ms) */
+export const QR_TOKEN_TTL_MS = 60_000;
+
+/** Grace period — previous token still valid during rotation (ms) */
+export const QR_TOKEN_GRACE_MS = 90_000;
+
+/** Length of the short code in QR URL path (chars) */
+export const SHORT_CODE_LENGTH = 6;
+
+// ============================================================================
+// QR Rate Limiting
+// ============================================================================
+
+/** Global rate limit for QR auth attempts across all IPs */
+export const QR_RATE_LIMIT_MAX = 30;
+
+/** QR rate limit reset window (ms) */
+export const QR_RATE_LIMIT_WINDOW_MS = 60_000;
+
+/** Per-IP rate limit for QR auth failures (separate from Basic Auth AUTH_FAILURE_MAX) */
+export const QR_AUTH_FAILURE_MAX = 10;
+
+// ============================================================================
+// Tunnel Process Lifecycle
+// ============================================================================
+
+/** Max time to wait for cloudflared URL before timeout (ms) */
+export const URL_TIMEOUT_MS = 30_000;
+
+/** Restart delay after unexpected tunnel exit (ms) */
+export const RESTART_DELAY_MS = 5_000;
+
+/** SIGTERM → SIGKILL escalation timeout (ms) */
+export const FORCE_KILL_MS = 5_000;

src/file-stream-manager.ts

@@ -16,6 +16,8 @@ import { existsSync, statSync, realpathSync } from 'node:fs';
 import { resolve, relative, isAbsolute } from 'node:path';
 import { homedir } from 'node:os';
 import { EventEmitter } from 'node:events';
+import { getErrorMessage } from './types.js';
+import { CLEANUP_CHECK_INTERVAL_MS, INACTIVITY_TIMEOUT_MS } from './config/server-timing.js';
 
 // ========== Configuration Constants ==========
 
@@ -39,7 +41,7 @@ const MAX_STREAMS_PER_SESSION = 5;
  * Inactivity timeout for streams (5 minutes).
  * Streams with no data for this long will be auto-closed.
  */
-const STREAM_INACTIVITY_TIMEOUT_MS = 5 * 60 * 1000;
+const STREAM_INACTIVITY_TIMEOUT_MS = INACTIVITY_TIMEOUT_MS;
 
 // ========== Types ==========
 
@@ -129,7 +131,7 @@ export class FileStreamManager extends EventEmitter {
   constructor() {
     super();
     // Start cleanup timer for inactive streams
-    this.cleanupTimer = setInterval(() => this.cleanupInactiveStreams(), 60 * 1000);
+    this.cleanupTimer = setInterval(() => this.cleanupInactiveStreams(), CLEANUP_CHECK_INTERVAL_MS);
   }
 
   // ========== Public Methods ==========
@@ -171,10 +173,7 @@ export class FileStreamManager extends EventEmitter {
       }
     } catch (err) {
       const errorCode = err instanceof Error && 'code' in err ? (err as NodeJS.ErrnoException).code : 'UNKNOWN';
-      console.warn(
-        `[FileStreamManager] Failed to stat file "${absolutePath}" (${errorCode}):`,
-        err instanceof Error ? err.message : String(err)
-      );
+      console.warn(`[FileStreamManager] Failed to stat file "${absolutePath}" (${errorCode}):`, getErrorMessage(err));
       return { success: false, error: 'File not found or not accessible' };
     }
 

src/hooks-config.ts

@@ -1,11 +1,26 @@
 /**
- * @fileoverview Claude Code hooks configuration generator
+ * @fileoverview Claude Code hooks configuration generator.
  *
- * Generates .claude/settings.local.json with hook definitions that POST
- * to Codeman's /api/hook-event endpoint when Claude Code fires
- * notification or stop hooks. Uses $CODEMAN_API_URL and
- * $CODEMAN_SESSION_ID env vars (set on every managed session) so the
- * config is static per case directory.
+ * Generates `.claude/settings.local.json` with hook definitions that POST
+ * to Codeman's `/api/hook-event` endpoint when Claude Code fires hooks.
+ * Uses `$CODEMAN_API_URL` and `$CODEMAN_SESSION_ID` env vars (set on every
+ * managed session) so the config is static per case directory.
+ *
+ * Key exports:
+ * - `generateHooksConfig()` — returns hooks object for settings.local.json
+ * - `writeHooksConfig(casePath)` — writes hooks + env config to disk
+ * - `updateCaseEnvVars(casePath, envVars)` — merges env vars into settings
+ *
+ * Hook events generated: `idle_prompt`, `permission_prompt`, `elicitation_dialog`,
+ * `stop`, `teammate_idle`, `task_completed`
+ *
+ * Hook categories: `Notification` (3 matchers), `Stop` (1), `TeammateIdle` (1),
+ * `TaskCompleted` (1)
+ *
+ * @dependencies types (HookEventType), config/auth-config (HOOK_TIMEOUT_MS)
+ * @consumedby web/server (session creation), session-cli-builder (env setup)
+ *
+ * @module hooks-config
  */
 
 import { existsSync } from 'node:fs';
@@ -13,6 +28,7 @@ import { readFile, writeFile, mkdir } from 'node:fs/promises';
 import { join } from 'node:path';
 
 import type { HookEventType } from './types.js';
+import { HOOK_TIMEOUT_MS } from './config/auth-config.js';
 
 /**
  * Generates the hooks section for .claude/settings.local.json
@@ -38,30 +54,30 @@ export function generateHooksConfig(): { hooks: Record<string, unknown[]> } {
       Notification: [
         {
           matcher: 'idle_prompt',
-          hooks: [{ type: 'command', command: curlCmd('idle_prompt'), timeout: 10000 }],
+          hooks: [{ type: 'command', command: curlCmd('idle_prompt'), timeout: HOOK_TIMEOUT_MS }],
         },
         {
           matcher: 'permission_prompt',
-          hooks: [{ type: 'command', command: curlCmd('permission_prompt'), timeout: 10000 }],
+          hooks: [{ type: 'command', command: curlCmd('permission_prompt'), timeout: HOOK_TIMEOUT_MS }],
         },
         {
           matcher: 'elicitation_dialog',
-          hooks: [{ type: 'command', command: curlCmd('elicitation_dialog'), timeout: 10000 }],
+          hooks: [{ type: 'command', command: curlCmd('elicitation_dialog'), timeout: HOOK_TIMEOUT_MS }],
         },
       ],
       Stop: [
         {
-          hooks: [{ type: 'command', command: curlCmd('stop'), timeout: 10000 }],
+          hooks: [{ type: 'command', command: curlCmd('stop'), timeout: HOOK_TIMEOUT_MS }],
         },
       ],
       TeammateIdle: [
         {
-          hooks: [{ type: 'command', command: curlCmd('teammate_idle'), timeout: 10000 }],
+          hooks: [{ type: 'command', command: curlCmd('teammate_idle'), timeout: HOOK_TIMEOUT_MS }],
         },
       ],
       TaskCompleted: [
         {
-          hooks: [{ type: 'command', command: curlCmd('task_completed'), timeout: 10000 }],
+          hooks: [{ type: 'command', command: curlCmd('task_completed'), timeout: HOOK_TIMEOUT_MS }],
         },
       ],
     },
@@ -100,6 +116,34 @@ export async function updateCaseEnvVars(casePath: string, envVars: Record<string
   await writeFile(settingsPath, JSON.stringify(existing, null, 2) + '\n');
 }
 
+/**
+ * Updates the `model` field in .claude/settings.local.json for the given case path.
+ * Pass a non-empty string to set, or empty/null to remove.
+ */
+export async function updateCaseModel(casePath: string, model: string | null): Promise<void> {
+  const claudeDir = join(casePath, '.claude');
+  if (!existsSync(claudeDir)) {
+    await mkdir(claudeDir, { recursive: true });
+  }
+
+  const settingsPath = join(claudeDir, 'settings.local.json');
+  let existing: Record<string, unknown> = {};
+
+  try {
+    existing = JSON.parse(await readFile(settingsPath, 'utf-8'));
+  } catch {
+    existing = {};
+  }
+
+  if (model) {
+    existing.model = model;
+  } else {
+    delete existing.model;
+  }
+
+  await writeFile(settingsPath, JSON.stringify(existing, null, 2) + '\n');
+}
+
 /**
  * Writes hooks config to .claude/settings.local.json in the given case path.
  * Merges with existing file content, only touching the `hooks` key.

src/image-watcher.ts

@@ -13,6 +13,7 @@ import { watch, type FSWatcher } from 'chokidar';
 import { basename, extname, relative } from 'node:path';
 import { statSync } from 'node:fs';
 import type { ImageDetectedEvent } from './types.js';
+import { KeyedDebouncer } from './utils/index.js';
 
 // ========== Types ==========
 
@@ -65,11 +66,11 @@ export class ImageWatcher extends EventEmitter {
   /** Map of sessionId -> working directory path */
   private sessionDirs = new Map<string, string>();
 
-  /** Debounce timers for rapid image creation (keyed by filePath) */
-  private debounceTimers = new Map<string, NodeJS.Timeout>();
+  /** Per-file debouncer for rapid image creation */
+  private fileDeb = new KeyedDebouncer(DEBOUNCE_DELAY_MS);
 
-  /** Track which session owns each debounce timer (for cleanup) */
-  private timerToSession = new Map<string, string>();
+  /** Track which session owns each debounced file (for cleanup) */
+  private fileToSession = new Map<string, string>();
 
   /** Per-session burst tracking: sessionId -> { count, windowStart } */
   private burstTrackers = new Map<string, { count: number; windowStart: number }>();
@@ -118,11 +119,8 @@ export class ImageWatcher extends EventEmitter {
     this.sessionDirs.clear();
 
     // Clear all debounce timers
-    for (const timer of this.debounceTimers.values()) {
-      clearTimeout(timer);
-    }
-    this.debounceTimers.clear();
-    this.timerToSession.clear();
+    this.fileDeb.dispose();
+    this.fileToSession.clear();
     this.burstTrackers.clear();
   }
 
@@ -212,20 +210,15 @@ export class ImageWatcher extends EventEmitter {
     this.sessionDirs.delete(sessionId);
 
     // Clear any pending debounce timers for this session
-    // Collect keys first to avoid iterator invalidation during deletion
-    const toDelete: string[] = [];
-    for (const [filePath, ownerId] of this.timerToSession) {
+    const toCancel: string[] = [];
+    for (const [filePath, ownerId] of this.fileToSession) {
       if (ownerId === sessionId) {
-        toDelete.push(filePath);
+        toCancel.push(filePath);
       }
     }
-    for (const filePath of toDelete) {
-      const timer = this.debounceTimers.get(filePath);
-      if (timer) {
-        clearTimeout(timer);
-        this.debounceTimers.delete(filePath);
-      }
-      this.timerToSession.delete(filePath);
+    for (const filePath of toCancel) {
+      this.fileDeb.cancelKey(filePath);
+      this.fileToSession.delete(filePath);
     }
     this.burstTrackers.delete(sessionId);
   }
@@ -269,22 +262,15 @@ export class ImageWatcher extends EventEmitter {
     }
 
     // Debounce rapid file creation (e.g., multiple screenshots quickly)
-    const existingTimer = this.debounceTimers.get(filePath);
-    if (existingTimer) {
-      clearTimeout(existingTimer);
-    }
-
-    const timer = setTimeout(() => {
-      this.debounceTimers.delete(filePath);
-      this.timerToSession.delete(filePath);
+    this.fileDeb.schedule(filePath, () => {
+      this.fileToSession.delete(filePath);
       this.emitImageDetected(sessionId, filePath);
       // Increment burst count on actual emission (not on detection)
       const b = this.burstTrackers.get(sessionId);
       if (b) b.count++;
-    }, DEBOUNCE_DELAY_MS);
+    });
 
-    this.debounceTimers.set(filePath, timer);
-    this.timerToSession.set(filePath, sessionId);
+    this.fileToSession.set(filePath, sessionId);
   }
 
   /**

src/index.ts

@@ -14,10 +14,10 @@ import { program } from './cli.js';
 // In web mode, we should NOT exit on transient errors — log and continue
 const isWebMode = process.argv.includes('web');
 
+import { MAX_CONSECUTIVE_ERRORS, ERROR_RESET_MS } from './config/server-timing.js';
+
 // Track consecutive unhandled errors in web mode — restart after too many
 let consecutiveErrors = 0;
-const MAX_CONSECUTIVE_ERRORS = 5;
-const ERROR_RESET_MS = 60000; // Reset counter after 1 minute of no errors
 let errorResetTimer: ReturnType<typeof setTimeout> | null = null;
 
 function trackError(): void {

src/mux-interface.ts

@@ -61,6 +61,8 @@ export interface CreateSessionOptions {
   claudeMode?: ClaudeMode;
   allowedTools?: string;
   openCodeConfig?: OpenCodeConfig;
+  /** When restoring after reboot, resume a previous Claude conversation by its session ID */
+  resumeSessionId?: string;
 }
 
 /** Options for respawning a dead pane. */
@@ -73,6 +75,8 @@ export interface RespawnPaneOptions {
   claudeMode?: ClaudeMode;
   allowedTools?: string;
   openCodeConfig?: OpenCodeConfig;
+  /** Resume a previous Claude conversation when respawning */
+  resumeSessionId?: string;
 }
 
 /**

src/orchestrator-loop.ts

@@ -0,0 +1,993 @@
+/**
+ * @fileoverview Orchestrator Loop — phased plan execution with team agents.
+ *
+ * State machine that generates plans from user goals, executes them
+ * phase-by-phase with verification gates, and adapts on failure.
+ *
+ * States: idle → planning → approval → executing → verifying → (replanning) → completed/failed
+ *
+ * Key exports:
+ * - `OrchestratorLoop` class — main engine, extends EventEmitter
+ * - `OrchestratorLoopEvents` interface — typed event map
+ *
+ * Lifecycle: `start(goal)` → plan → approve → execute phases → verify → complete
+ *
+ * @dependencies orchestrator-planner (plan generation), orchestrator-verifier (phase verification),
+ *   session-manager (sessions), task-queue (task execution), state-store (persistence),
+ *   prompts/orchestrator (prompt templates)
+ * @consumedby web/server (orchestrator routes, SSE)
+ * @emits stateChanged, planReady, phaseStarted, phaseCompleted, phaseFailed,
+ *   taskAssigned, taskCompleted, taskFailed, verificationResult, completed, error
+ * @persistence Orchestrator state saved to `~/.codeman/state.json` (orchestrator key)
+ *
+ * @module orchestrator-loop
+ */
+
+import { EventEmitter } from 'node:events';
+import { getSessionManager, SessionManager } from './session-manager.js';
+import { getTaskQueue, TaskQueue } from './task-queue.js';
+import { getStore, StateStore } from './state-store.js';
+import { OrchestratorPlanner } from './orchestrator-planner.js';
+import { OrchestratorVerifier } from './orchestrator-verifier.js';
+import { PHASE_EXECUTION_PROMPT, REPLAN_PROMPT, SINGLE_TASK_PROMPT, TEAM_LEAD_PROMPT } from './prompts/index.js';
+import type { TerminalMultiplexer } from './mux-interface.js';
+import type { CreateTaskOptions } from './task.js';
+import {
+  type OrchestratorState,
+  type OrchestratorPlan,
+  type OrchestratorPhase,
+  type OrchestratorTask,
+  type OrchestratorConfig,
+  type OrchestratorStats,
+  type OrchestratorPersistState,
+  type VerificationResult,
+  DEFAULT_ORCHESTRATOR_CONFIG,
+  createInitialOrchestratorStats,
+  getErrorMessage,
+} from './types.js';
+
+// ═══════════════════════════════════════════════════════════════
+// Constants
+// ═══════════════════════════════════════════════════════════════
+
+/** Poll interval for checking task completion within a phase (2 seconds) */
+const PHASE_POLL_INTERVAL_MS = 2000;
+
+/** Delay between phase completion and verification (1 second) */
+const POST_PHASE_DELAY_MS = 1000;
+
+// ═══════════════════════════════════════════════════════════════
+// Events
+// ═══════════════════════════════════════════════════════════════
+
+export interface OrchestratorLoopEvents {
+  stateChanged: (state: OrchestratorState, prevState: OrchestratorState) => void;
+  planProgress: (phase: string, detail: string) => void;
+  planReady: (plan: OrchestratorPlan) => void;
+  phaseStarted: (phase: OrchestratorPhase) => void;
+  phaseCompleted: (phase: OrchestratorPhase) => void;
+  phaseFailed: (phase: OrchestratorPhase, reason: string) => void;
+  taskAssigned: (task: OrchestratorTask, sessionId: string) => void;
+  taskCompleted: (task: OrchestratorTask) => void;
+  taskFailed: (task: OrchestratorTask, error: string) => void;
+  verificationResult: (phase: OrchestratorPhase, result: VerificationResult) => void;
+  completed: (stats: OrchestratorStats) => void;
+  error: (error: Error) => void;
+}
+
+// ═══════════════════════════════════════════════════════════════
+// OrchestratorLoop
+// ═══════════════════════════════════════════════════════════════
+
+export class OrchestratorLoop extends EventEmitter {
+  private _state: OrchestratorState = 'idle';
+  private plan: OrchestratorPlan | null = null;
+  private currentPhaseIndex = 0;
+  private config: OrchestratorConfig;
+  private stats: OrchestratorStats;
+  private startedAt: number | null = null;
+  private completedAt: number | null = null;
+
+  private workingDir: string;
+  private planner: OrchestratorPlanner;
+  private verifier: OrchestratorVerifier;
+  private sessionManager: SessionManager;
+  private taskQueue: TaskQueue;
+  private store: StateStore;
+
+  /** State before pause (to resume to correct state) */
+  private pausedState: OrchestratorState | null = null;
+
+  /** Phase poll timer for checking task completion */
+  private phasePollTimer: NodeJS.Timeout | null = null;
+
+  /** Phase-level timeout timer */
+  private phaseTimeoutTimer: NodeJS.Timeout | null = null;
+
+  /** Post-phase delay timer before verification */
+  private postPhaseTimer: NodeJS.Timeout | null = null;
+
+  /** Session completion listener (bound for cleanup) */
+  private sessionCompletionListener: ((sessionId: string, phrase: string) => void) | null = null;
+
+  /** Active sessions assigned to current phase */
+  private phaseSessionIds: Set<string> = new Set();
+
+  constructor(mux: TerminalMultiplexer, workingDir: string, config?: Partial<OrchestratorConfig>) {
+    super();
+    this.workingDir = workingDir;
+    this.config = { ...DEFAULT_ORCHESTRATOR_CONFIG, ...config };
+    this.stats = createInitialOrchestratorStats();
+    this.sessionManager = getSessionManager();
+    this.taskQueue = getTaskQueue();
+    this.store = getStore();
+    this.planner = new OrchestratorPlanner(mux, workingDir, this.config);
+    this.verifier = new OrchestratorVerifier(this.config);
+
+    // Restore state if crashed while running
+    this.restore();
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Public API — Lifecycle
+  // ═══════════════════════════════════════════════════════════════
+
+  /** Start orchestration with a goal. Transitions: idle → planning */
+  async start(goal: string): Promise<void> {
+    if (this._state !== 'idle' && this._state !== 'failed' && this._state !== 'completed') {
+      throw new Error(`Cannot start from state "${this._state}"`);
+    }
+
+    this.reset();
+    this.startedAt = Date.now();
+    this.setState('planning');
+
+    try {
+      const plan = await this.planner.generatePlan(goal, (phase, detail) => {
+        this.emit('planProgress', phase, detail);
+      });
+
+      if (this.currentState() !== 'planning') {
+        // Cancelled during planning
+        return;
+      }
+
+      this.plan = plan;
+      this.persist();
+
+      if (this.config.autoApprove) {
+        this.setState('executing');
+        await this.executeCurrentPhase();
+      } else {
+        this.setState('approval');
+        this.emit('planReady', plan);
+      }
+    } catch (err) {
+      this.handleError(err);
+    }
+  }
+
+  /** Approve the generated plan. Transitions: approval → executing */
+  async approve(): Promise<void> {
+    this.requireState('approval');
+    if (!this.plan) {
+      throw new Error('No plan to approve');
+    }
+
+    this.setState('executing');
+    await this.executeCurrentPhase();
+  }
+
+  /** Reject plan with feedback. Transitions: approval → planning (regenerate) */
+  async reject(feedback: string): Promise<void> {
+    this.requireState('approval');
+    if (!this.plan) {
+      throw new Error('No plan to reject');
+    }
+
+    const goal = this.plan.goal + '\n\nFeedback on previous plan: ' + feedback;
+    this.plan = null;
+    this.setState('planning');
+
+    try {
+      const plan = await this.planner.generatePlan(goal);
+
+      if ((this._state as OrchestratorState) !== 'planning') return;
+
+      this.plan = plan;
+      this.persist();
+      this.setState('approval');
+      this.emit('planReady', plan);
+    } catch (err) {
+      this.handleError(err);
+    }
+  }
+
+  /** Pause execution. Saves current state. */
+  pause(): void {
+    if (this._state === 'idle' || this._state === 'paused' || this._state === 'completed' || this._state === 'failed') {
+      return;
+    }
+    this.pausedState = this._state;
+    this.clearPhasePoll();
+    this.cleanupTaskHandlers();
+    this.setState('paused');
+  }
+
+  /** Resume from pause. */
+  async resume(): Promise<void> {
+    if (this._state !== 'paused' || !this.pausedState) {
+      throw new Error('Not paused');
+    }
+
+    const resumeTo = this.pausedState;
+    this.pausedState = null;
+    this.setState(resumeTo);
+
+    // Re-enter the appropriate phase of execution
+    if (resumeTo === 'executing') {
+      await this.executeCurrentPhase();
+    } else if (resumeTo === 'verifying') {
+      await this.verifyCurrentPhase();
+    }
+  }
+
+  /** Stop everything and clean up. */
+  async stop(): Promise<void> {
+    this.clearPhasePoll();
+    this.cleanupTaskHandlers();
+    await this.planner.cancel();
+    this.setState('idle');
+    this.store.clearOrchestratorState();
+  }
+
+  /** Skip a specific phase. */
+  async skipPhase(phaseId: string): Promise<void> {
+    if (!this.plan) return;
+
+    const phase = this.plan.phases.find((p) => p.id === phaseId);
+    if (!phase) throw new Error(`Phase "${phaseId}" not found`);
+
+    phase.status = 'skipped';
+    phase.completedAt = Date.now();
+    this.persist();
+
+    // If this is the current phase, advance
+    if (this.plan.phases[this.currentPhaseIndex]?.id === phaseId) {
+      await this.advanceToNextPhase();
+    }
+  }
+
+  /** Retry a failed phase. */
+  async retryPhase(phaseId: string): Promise<void> {
+    if (!this.plan) return;
+    if (this._state !== 'executing' && this._state !== 'failed') {
+      throw new Error(`Cannot retry from state "${this._state}"`);
+    }
+
+    const phaseIndex = this.plan.phases.findIndex((p) => p.id === phaseId);
+    if (phaseIndex === -1) throw new Error(`Phase "${phaseId}" not found`);
+
+    const phase = this.plan.phases[phaseIndex];
+    phase.status = 'pending';
+    phase.attempts = 0;
+    for (const task of phase.tasks) {
+      task.status = 'pending';
+      task.error = null;
+      task.assignedSessionId = null;
+      task.queueTaskId = null;
+    }
+
+    this.currentPhaseIndex = phaseIndex;
+    this.setState('executing');
+    await this.executeCurrentPhase();
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Public API — Getters
+  // ═══════════════════════════════════════════════════════════════
+
+  get state(): OrchestratorState {
+    return this._state;
+  }
+
+  getPlan(): OrchestratorPlan | null {
+    return this.plan;
+  }
+
+  getCurrentPhase(): OrchestratorPhase | null {
+    if (!this.plan) return null;
+    return this.plan.phases[this.currentPhaseIndex] ?? null;
+  }
+
+  getStats(): OrchestratorStats {
+    return { ...this.stats };
+  }
+
+  getStatus(): OrchestratorPersistState {
+    return {
+      state: this._state,
+      plan: this.plan,
+      currentPhaseIndex: this.currentPhaseIndex,
+      startedAt: this.startedAt,
+      completedAt: this.completedAt,
+      config: this.config,
+      stats: this.stats,
+    };
+  }
+
+  isRunning(): boolean {
+    return this._state !== 'idle' && this._state !== 'completed' && this._state !== 'failed';
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Internal — Phase Execution
+  // ═══════════════════════════════════════════════════════════════
+
+  private async executeCurrentPhase(): Promise<void> {
+    if (!this.plan || this._state !== 'executing') return;
+
+    const phase = this.plan.phases[this.currentPhaseIndex];
+    if (!phase) {
+      // All phases done
+      await this.handleCompletion();
+      return;
+    }
+
+    // Skip already completed/skipped phases
+    if (phase.status === 'passed' || phase.status === 'skipped') {
+      await this.advanceToNextPhase();
+      return;
+    }
+
+    phase.status = 'executing';
+    phase.startedAt = Date.now();
+    phase.attempts++;
+    this.persist();
+    this.emit('phaseStarted', phase);
+
+    try {
+      await this.assignPhaseTasks(phase);
+      this.startPhasePoll(phase);
+    } catch (err) {
+      this.handlePhaseError(phase, getErrorMessage(err));
+    }
+  }
+
+  private async assignPhaseTasks(phase: OrchestratorPhase): Promise<void> {
+    // For team strategy, send a single comprehensive prompt to a lead session
+    if (phase.teamStrategy.type === 'team') {
+      await this.assignTeamPhase(phase);
+      return;
+    }
+
+    // For single/parallel strategy, add individual tasks to TaskQueue
+    for (const task of phase.tasks) {
+      if (task.status !== 'pending') continue;
+
+      const prompt = this.buildTaskPrompt(task, phase);
+      const taskOptions: CreateTaskOptions = {
+        prompt,
+        workingDir: this.workingDir,
+        priority: 100 - phase.order, // Earlier phases get higher priority
+        completionPhrase: task.completionPhrase,
+        timeoutMs: Math.min(task.timeoutMs, this.config.phaseTimeoutMs),
+      };
+
+      const queueTask = this.taskQueue.addTask(taskOptions);
+      task.queueTaskId = queueTask.id;
+      task.status = 'running';
+    }
+
+    this.persist();
+    this.setupTaskHandlers();
+
+    // Manually assign tasks to idle sessions
+    await this.assignQueuedTasksToSessions();
+  }
+
+  private async assignTeamPhase(phase: OrchestratorPhase): Promise<void> {
+    const teamConfig = phase.teamStrategy.type === 'team' ? phase.teamStrategy.config : null;
+    if (!teamConfig) return;
+
+    // Find or use an idle session
+    const sessions = this.sessionManager.getIdleSessions();
+    if (sessions.length === 0) {
+      throw new Error('No idle sessions available for team phase execution');
+    }
+
+    const session = sessions[0];
+    this.phaseSessionIds.add(session.id);
+
+    // Mark all tasks as running under this session
+    for (const task of phase.tasks) {
+      task.status = 'running';
+      task.assignedSessionId = session.id;
+    }
+
+    // Build and send the team lead prompt
+    const prompt = TEAM_LEAD_PROMPT.replace('{PHASE_NAME}', phase.name)
+      .replace('{TASK_LIST}', phase.tasks.map((t, i) => `${i + 1}. ${t.prompt}`).join('\n'))
+      .replace('{TEAMMATE_HINTS}', teamConfig.suggestedTeammates.map((h, i) => `${i + 1}. ${h}`).join('\n'))
+      .replace('{COMPLETION_PHRASE}', `${phase.id.toUpperCase()}_COMPLETE`);
+
+    // Create a TaskQueue task for the entire phase
+    const queueTask = this.taskQueue.addTask({
+      prompt,
+      workingDir: this.workingDir,
+      priority: 100 - phase.order,
+      completionPhrase: `${phase.id.toUpperCase()}_COMPLETE`,
+      timeoutMs: this.config.phaseTimeoutMs,
+    });
+
+    // Link all phase tasks to this single queue task
+    for (const task of phase.tasks) {
+      task.queueTaskId = queueTask.id;
+    }
+
+    this.persist();
+    this.setupTaskHandlers();
+
+    // Assign the task to the session
+    try {
+      queueTask.assign(session.id);
+      session.assignTask(queueTask.id);
+      this.taskQueue.updateTask(queueTask);
+      await session.sendInput(prompt);
+    } catch (err) {
+      queueTask.fail(getErrorMessage(err));
+      this.taskQueue.updateTask(queueTask);
+      throw err;
+    }
+  }
+
+  private async assignQueuedTasksToSessions(): Promise<void> {
+    const idleSessions = this.sessionManager.getIdleSessions();
+    const maxSessions =
+      this.getCurrentPhase()?.teamStrategy.type === 'parallel'
+        ? (this.getCurrentPhase()?.teamStrategy as { type: 'parallel'; maxSessions: number }).maxSessions
+        : 1;
+
+    const sessionsToUse = idleSessions.slice(0, maxSessions);
+
+    for (const session of sessionsToUse) {
+      const task = this.taskQueue.next();
+      if (!task) break;
+
+      try {
+        task.assign(session.id);
+        session.assignTask(task.id);
+        this.taskQueue.updateTask(task);
+        await session.sendInput(task.prompt);
+
+        this.phaseSessionIds.add(session.id);
+
+        // Find the orchestrator task linked to this queue task
+        const orchTask = this.findOrchestratorTaskByQueueId(task.id);
+        if (orchTask) {
+          orchTask.assignedSessionId = session.id;
+          orchTask.startedAt = Date.now();
+          this.emit('taskAssigned', orchTask, session.id);
+        }
+      } catch (err) {
+        task.fail(getErrorMessage(err));
+        session.clearTask();
+        this.taskQueue.updateTask(task);
+      }
+    }
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Internal — Task Completion Tracking
+  // ═══════════════════════════════════════════════════════════════
+
+  private setupTaskHandlers(): void {
+    this.cleanupTaskHandlers();
+
+    this.sessionCompletionListener = (_sessionId: string, _phrase: string) => {
+      // Session completion — check if it's related to our phase tasks
+      this.checkPhaseCompletion();
+    };
+
+    this.sessionManager.on('sessionCompletion', this.sessionCompletionListener);
+  }
+
+  private cleanupTaskHandlers(): void {
+    if (this.sessionCompletionListener) {
+      this.sessionManager.off('sessionCompletion', this.sessionCompletionListener);
+      this.sessionCompletionListener = null;
+    }
+  }
+
+  private _finalizeTask(queueTaskId: string, status: 'completed' | 'failed', error?: string): OrchestratorTask | null {
+    const orchTask = this.findOrchestratorTaskByQueueId(queueTaskId);
+    if (!orchTask) return null;
+
+    orchTask.status = status;
+    if (status === 'completed') {
+      orchTask.completedAt = Date.now();
+      this.stats.totalTasksCompleted++;
+    } else {
+      orchTask.error = error ?? null;
+      this.stats.totalTasksFailed++;
+    }
+    this.persist();
+    return orchTask;
+  }
+
+  private handleTaskCompleted(queueTaskId: string): void {
+    const orchTask = this._finalizeTask(queueTaskId, 'completed');
+    if (!orchTask) return;
+
+    this.emit('taskCompleted', orchTask);
+    this.checkPhaseCompletion();
+  }
+
+  private handleTaskFailed(queueTaskId: string, error: string): void {
+    const orchTask = this._finalizeTask(queueTaskId, 'failed', error);
+    if (!orchTask) return;
+
+    this.emit('taskFailed', orchTask, error);
+
+    // Check if we should retry the task or fail the phase
+    if (orchTask.retries < 2) {
+      orchTask.retries++;
+      orchTask.status = 'pending';
+      orchTask.error = null;
+      orchTask.queueTaskId = null;
+      // Will be re-queued on next poll
+    } else {
+      this.checkPhaseCompletion();
+    }
+  }
+
+  private startPhasePoll(phase: OrchestratorPhase): void {
+    this.clearPhasePoll();
+    this.phasePollTimer = setInterval(() => {
+      if (this._state !== 'executing') {
+        this.clearPhasePoll();
+        return;
+      }
+      this.pollPhaseStatus(phase);
+    }, PHASE_POLL_INTERVAL_MS);
+
+    // Phase-level timeout — fail the phase if it exceeds the configured timeout
+    this.phaseTimeoutTimer = setTimeout(() => {
+      if (this._state === 'executing' && phase.status === 'executing') {
+        console.warn(`[Orchestrator] Phase "${phase.name}" timed out after ${this.config.phaseTimeoutMs}ms`);
+        this.handlePhaseError(phase, `Phase timed out after ${Math.round(this.config.phaseTimeoutMs / 60000)} minutes`);
+      }
+    }, this.config.phaseTimeoutMs);
+  }
+
+  private _clearTimer(
+    timerKey: 'phasePollTimer' | 'phaseTimeoutTimer' | 'postPhaseTimer',
+    clearFn: typeof clearInterval | typeof clearTimeout
+  ): void {
+    if (this[timerKey]) {
+      clearFn(this[timerKey]);
+      this[timerKey] = null;
+    }
+  }
+
+  private clearPhasePoll(): void {
+    this._clearTimer('phasePollTimer', clearInterval);
+    this._clearTimer('phaseTimeoutTimer', clearTimeout);
+    this._clearTimer('postPhaseTimer', clearTimeout);
+  }
+
+  private pollPhaseStatus(phase: OrchestratorPhase): void {
+    // Check for queued tasks that need assignment
+    const pendingTasks = phase.tasks.filter((t) => t.status === 'pending' && !t.queueTaskId);
+    if (pendingTasks.length > 0) {
+      // Re-queue pending tasks
+      for (const task of pendingTasks) {
+        const prompt = this.buildTaskPrompt(task, phase);
+        const queueTask = this.taskQueue.addTask({
+          prompt,
+          workingDir: this.workingDir,
+          priority: 100 - phase.order,
+          completionPhrase: task.completionPhrase,
+          timeoutMs: Math.min(task.timeoutMs, this.config.phaseTimeoutMs),
+        });
+        task.queueTaskId = queueTask.id;
+        task.status = 'running';
+      }
+      this.assignQueuedTasksToSessions().catch(() => {}); // Best effort
+    }
+
+    // Check completion status of queue tasks
+    for (const task of phase.tasks) {
+      if (task.status === 'running' && task.queueTaskId) {
+        const queueTask = this.taskQueue.getTask(task.queueTaskId);
+        if (queueTask) {
+          if (queueTask.isCompleted()) {
+            this.handleTaskCompleted(task.queueTaskId);
+          } else if (queueTask.isFailed()) {
+            this.handleTaskFailed(task.queueTaskId, queueTask.error || 'Task failed');
+          }
+        }
+      }
+    }
+
+    this.checkPhaseCompletion();
+  }
+
+  private checkPhaseCompletion(): void {
+    if (this._state !== 'executing') return;
+
+    const phase = this.getCurrentPhase();
+    if (!phase) return;
+
+    const allDone = phase.tasks.every((t) => t.status === 'completed' || t.status === 'failed');
+    if (!allDone) return;
+
+    const anyFailed = phase.tasks.some((t) => t.status === 'failed');
+
+    this.clearPhasePoll();
+
+    if (anyFailed) {
+      // Phase has failed tasks
+      this.handlePhaseError(phase, 'One or more tasks failed');
+    } else {
+      // All tasks completed — run verification after brief delay
+      this.postPhaseTimer = setTimeout(() => {
+        this.postPhaseTimer = null;
+        this.verifyCurrentPhase().catch((err) => this.handleError(err));
+      }, POST_PHASE_DELAY_MS);
+    }
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Internal — Verification
+  // ═══════════════════════════════════════════════════════════════
+
+  private async verifyCurrentPhase(): Promise<void> {
+    if (!this.plan) return;
+
+    const phase = this.plan.phases[this.currentPhaseIndex];
+    if (!phase) return;
+
+    // Skip verification if no criteria defined
+    if (phase.verificationCriteria.length === 0 && phase.testCommands.length === 0) {
+      phase.status = 'passed';
+      phase.completedAt = Date.now();
+      phase.durationMs = phase.startedAt ? Date.now() - phase.startedAt : null;
+      this.stats.phasesCompleted++;
+      this.persist();
+      this.emit('phaseCompleted', phase);
+      await this.advanceToNextPhase();
+      return;
+    }
+
+    this.setState('verifying');
+
+    // Get a session for verification — wait briefly for sessions to become idle
+    let sessions = this.sessionManager.getIdleSessions();
+    if (sessions.length === 0) {
+      // Wait up to 10s for a session to become idle
+      await new Promise((resolve) => setTimeout(resolve, 10_000));
+      sessions = this.sessionManager.getIdleSessions();
+    }
+    if (sessions.length === 0) {
+      // Still no sessions — log warning and skip verification (don't silently pass)
+      console.warn('[Orchestrator] No idle sessions for verification — skipping (marking passed with warning)');
+      phase.status = 'passed';
+      phase.completedAt = Date.now();
+      phase.durationMs = phase.startedAt ? Date.now() - phase.startedAt : null;
+      this.stats.phasesCompleted++;
+      this.persist();
+      this.emit('phaseCompleted', phase);
+      this.setState('executing');
+      await this.advanceToNextPhase();
+      return;
+    }
+
+    try {
+      const result = await this.verifier.verifyPhase(phase, sessions[0]);
+      this.emit('verificationResult', phase, result);
+
+      if (result.passed) {
+        phase.status = 'passed';
+        phase.completedAt = Date.now();
+        phase.durationMs = phase.startedAt ? Date.now() - phase.startedAt : null;
+        this.stats.phasesCompleted++;
+        this.persist();
+        this.emit('phaseCompleted', phase);
+        this.setState('executing');
+        await this.advanceToNextPhase();
+      } else {
+        // Verification failed — attempt replan
+        await this.handleVerificationFailure(phase, result);
+      }
+    } catch (err) {
+      // Verification error — treat as pass (don't block on verification bugs)
+      console.warn('[Orchestrator] Verification error, treating as pass:', err);
+      phase.status = 'passed';
+      phase.completedAt = Date.now();
+      phase.durationMs = phase.startedAt ? Date.now() - phase.startedAt : null;
+      this.stats.phasesCompleted++;
+      this.persist();
+      this.emit('phaseCompleted', phase);
+      this.setState('executing');
+      await this.advanceToNextPhase();
+    }
+  }
+
+  private async handleVerificationFailure(phase: OrchestratorPhase, result: VerificationResult): Promise<void> {
+    if (phase.attempts >= phase.maxAttempts) {
+      // Max retries exceeded
+      phase.status = 'failed';
+      phase.completedAt = Date.now();
+      phase.durationMs = phase.startedAt ? Date.now() - phase.startedAt : null;
+      this.stats.phasesFailed++;
+      this.persist();
+      this.emit('phaseFailed', phase, `Verification failed after ${phase.attempts} attempts: ${result.summary}`);
+      this.setState('failed');
+      return;
+    }
+
+    // Replan and retry
+    this.stats.replanCount++;
+    this.setState('replanning');
+
+    try {
+      await this.replanPhase(phase, result);
+      // Reset task states for retry
+      for (const task of phase.tasks) {
+        task.status = 'pending';
+        task.error = null;
+        task.assignedSessionId = null;
+        task.queueTaskId = null;
+        task.completedAt = null;
+        task.startedAt = null;
+      }
+      phase.status = 'pending';
+      phase.startedAt = null;
+      this.persist();
+
+      this.setState('executing');
+      await this.executeCurrentPhase();
+    } catch (err) {
+      this.handleError(err);
+    }
+  }
+
+  private async replanPhase(phase: OrchestratorPhase, result: VerificationResult): Promise<void> {
+    const completionPhrase = phase.tasks[0]?.completionPhrase || `${phase.id.toUpperCase()}_FIXED`;
+
+    const prompt = REPLAN_PROMPT.replace('{PHASE_NAME}', phase.name)
+      .replace('{ATTEMPT_NUMBER}', String(phase.attempts))
+      .replace('{MAX_ATTEMPTS}', String(phase.maxAttempts))
+      .replace('{FAILURE_SUMMARY}', result.summary)
+      .replace('{SUGGESTIONS}', result.suggestions.join('\n'))
+      .replace('{ORIGINAL_TASKS}', phase.tasks.map((t, i) => `${i + 1}. ${t.prompt}`).join('\n'))
+      .replace('{COMPLETION_PHRASE}', completionPhrase);
+
+    // Create a tracked queue task for the replan (so completion is detected)
+    const queueTask = this.taskQueue.addTask({
+      prompt,
+      workingDir: this.workingDir,
+      priority: 100,
+      completionPhrase,
+      timeoutMs: this.config.phaseTimeoutMs,
+    });
+
+    // Link to first phase task for tracking
+    if (phase.tasks[0]) {
+      phase.tasks[0].queueTaskId = queueTask.id;
+      phase.tasks[0].status = 'running';
+    }
+
+    this.persist();
+
+    // Set up handlers so task completion is tracked
+    this.setupTaskHandlers();
+
+    // Assign to a session
+    const sessions = this.sessionManager.getIdleSessions();
+    if (sessions.length === 0) {
+      console.warn('[Orchestrator] No idle sessions for replan — task queued, will pick up on next poll');
+      // Start polling so the task gets assigned when a session becomes idle
+      this.startPhasePoll(phase);
+      return;
+    }
+
+    try {
+      queueTask.assign(sessions[0].id);
+      sessions[0].assignTask(queueTask.id);
+      this.taskQueue.updateTask(queueTask);
+      await sessions[0].sendInput(prompt);
+    } catch (err) {
+      queueTask.fail(getErrorMessage(err));
+      this.taskQueue.updateTask(queueTask);
+    }
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Internal — State Machine
+  // ═══════════════════════════════════════════════════════════════
+
+  /** Read current state (bypasses TypeScript narrowing from guards) */
+  private currentState(): OrchestratorState {
+    return this._state;
+  }
+
+  /** Assert state matches expected or throw */
+  private requireState(...expected: OrchestratorState[]): void {
+    if (!expected.includes(this._state)) {
+      throw new Error(`Expected state "${expected.join('|')}", got "${this._state}"`);
+    }
+  }
+
+  private setState(newState: OrchestratorState): void {
+    const prev = this._state;
+    if (prev === newState) return;
+    this._state = newState;
+    this.persist();
+    this.emit('stateChanged', newState, prev);
+  }
+
+  private async advanceToNextPhase(): Promise<void> {
+    this.currentPhaseIndex++;
+    this.phaseSessionIds.clear();
+    this.persist();
+
+    if (!this.plan || this.currentPhaseIndex >= this.plan.phases.length) {
+      await this.handleCompletion();
+    } else {
+      // Compact between phases if configured
+      if (this.config.compactBetweenPhases) {
+        const sessions = this.sessionManager.getIdleSessions();
+        for (const session of sessions) {
+          try {
+            await session.writeViaMux('/compact');
+          } catch {
+            // Best effort
+          }
+        }
+        // Brief delay for compact to take effect
+        await new Promise((resolve) => setTimeout(resolve, 2000));
+      }
+
+      await this.executeCurrentPhase();
+    }
+  }
+
+  private async handleCompletion(): Promise<void> {
+    this.completedAt = Date.now();
+    this.stats.totalDurationMs = this.startedAt ? this.completedAt - this.startedAt : 0;
+    this.clearPhasePoll();
+    this.cleanupTaskHandlers();
+    this.setState('completed');
+    this.emit('completed', this.stats);
+  }
+
+  private handlePhaseError(phase: OrchestratorPhase, error: string): void {
+    if (phase.attempts >= phase.maxAttempts) {
+      phase.status = 'failed';
+      phase.completedAt = Date.now();
+      phase.durationMs = phase.startedAt ? Date.now() - phase.startedAt : null;
+      this.stats.phasesFailed++;
+      this.persist();
+      this.emit('phaseFailed', phase, error);
+      this.setState('failed');
+    } else {
+      // Retry the phase
+      for (const task of phase.tasks) {
+        if (task.status === 'failed') {
+          task.status = 'pending';
+          task.error = null;
+          task.queueTaskId = null;
+          task.assignedSessionId = null;
+        }
+      }
+      phase.status = 'pending';
+      this.persist();
+      this.executeCurrentPhase().catch((err) => this.handleError(err));
+    }
+  }
+
+  private handleError(err: unknown): void {
+    const error = err instanceof Error ? err : new Error(getErrorMessage(err));
+    console.error('[Orchestrator] Error:', error.message);
+    this.setState('failed');
+    this.emit('error', error);
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Internal — Persistence
+  // ═══════════════════════════════════════════════════════════════
+
+  private persist(): void {
+    this.store.setOrchestratorState(this.getStatus());
+  }
+
+  private restore(): void {
+    const saved = this.store.getOrchestratorState();
+    if (!saved) return;
+
+    // If we crashed while running, reset to failed
+    if (saved.state === 'executing' || saved.state === 'verifying' || saved.state === 'replanning') {
+      this._state = 'failed';
+      this.plan = saved.plan;
+      this.currentPhaseIndex = saved.currentPhaseIndex;
+      this.startedAt = saved.startedAt;
+      this.config = saved.config;
+      this.stats = saved.stats;
+      this.store.setOrchestratorState({ ...saved, state: 'failed' });
+    } else if (saved.state === 'planning' || saved.state === 'approval') {
+      // Planning/approval — reset to idle (plan is lost)
+      this.store.clearOrchestratorState();
+    } else if (saved.state === 'completed' || saved.state === 'failed') {
+      // Preserve completed/failed state for UI display
+      this._state = saved.state;
+      this.plan = saved.plan;
+      this.currentPhaseIndex = saved.currentPhaseIndex;
+      this.startedAt = saved.startedAt;
+      this.completedAt = saved.completedAt;
+      this.config = saved.config;
+      this.stats = saved.stats;
+    }
+  }
+
+  private reset(): void {
+    this._state = 'idle';
+    this.plan = null;
+    this.currentPhaseIndex = 0;
+    this.startedAt = null;
+    this.completedAt = null;
+    this.stats = createInitialOrchestratorStats();
+    this.pausedState = null;
+    this.phaseSessionIds.clear();
+    this.clearPhasePoll();
+    this.cleanupTaskHandlers();
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Internal — Helpers
+  // ═══════════════════════════════════════════════════════════════
+
+  private buildTaskPrompt(task: OrchestratorTask, phase: OrchestratorPhase): string {
+    if (phase.tasks.length === 1) {
+      // Single task — use simpler prompt
+      const completedPhases = this.getCompletedPhasesSummary();
+      return SINGLE_TASK_PROMPT.replace('{TASK}', task.prompt)
+        .replace('{GOAL}', this.plan?.goal || '')
+        .replace('{CONTEXT}', completedPhases ? `Previous phases completed: ${completedPhases}` : '')
+        .replace('{COMPLETION_PHRASE}', task.completionPhrase);
+    }
+
+    // Multi-task phase — use full prompt
+    return PHASE_EXECUTION_PROMPT.replace('{PHASE_NAME}', phase.name)
+      .replace('{GOAL}', this.plan?.goal || '')
+      .replace('{COMPLETED_PHASES}', this.getCompletedPhasesSummary() || 'None yet')
+      .replace('{TASK_LIST}', phase.tasks.map((t, i) => `${i + 1}. ${t.prompt}`).join('\n'))
+      .replace('{VERIFICATION_CRITERIA}', phase.verificationCriteria.join('\n') || 'No specific criteria')
+      .replace('{COMPLETION_PHRASE}', task.completionPhrase);
+  }
+
+  private getCompletedPhasesSummary(): string {
+    if (!this.plan) return '';
+    return this.plan.phases
+      .filter((p) => p.status === 'passed' || p.status === 'skipped')
+      .map((p) => `${p.name}: ${p.status}`)
+      .join(', ');
+  }
+
+  private findOrchestratorTaskByQueueId(queueTaskId: string): OrchestratorTask | null {
+    if (!this.plan) return null;
+    for (const phase of this.plan.phases) {
+      for (const task of phase.tasks) {
+        if (task.queueTaskId === queueTaskId) return task;
+      }
+    }
+    return null;
+  }
+
+  /** Clean up resources when the loop is being destroyed. */
+  destroy(): void {
+    this.clearPhasePoll();
+    this.cleanupTaskHandlers();
+  }
+}

src/orchestrator-planner.ts

@@ -0,0 +1,412 @@
+/**
+ * @fileoverview Orchestrator plan generation — converts goals into phased plans.
+ *
+ * Wraps PlanOrchestrator for AI-powered plan generation, then groups the
+ * resulting PlanItems into sequential phases with team strategies and
+ * verification criteria.
+ *
+ * Phase grouping algorithm:
+ * 1. Topological sort by dependencies (Kahn's algorithm)
+ * 2. Group into dependency layers
+ * 3. Sub-group by TDD phase within layers
+ * 4. Merge small adjacent phases
+ * 5. Assign team strategies based on parallelism potential
+ *
+ * Key exports:
+ * - `OrchestratorPlanner` class — plan generation + phase grouping
+ *
+ * @dependencies plan-orchestrator (AI plan generation), types (OrchestratorPlan, PlanItem)
+ * @consumedby orchestrator-loop
+ *
+ * @module orchestrator-planner
+ */
+
+import { v4 as uuidv4 } from 'uuid';
+import { PlanOrchestrator, type DetailedPlanResult, type ProgressCallback } from './plan-orchestrator.js';
+import type { TerminalMultiplexer } from './mux-interface.js';
+import type {
+  PlanItem,
+  TddPhase,
+  OrchestratorPlan,
+  OrchestratorPhase,
+  OrchestratorTask,
+  OrchestratorConfig,
+  TeamStrategy,
+  PhaseStatus,
+} from './types.js';
+
+// ═══════════════════════════════════════════════════════════════
+// Constants
+// ═══════════════════════════════════════════════════════════════
+
+/** Maximum number of phases (prevents runaway plans) */
+const MAX_PHASES = 10;
+
+/** Maximum total tasks across all phases */
+const MAX_TOTAL_TASKS = 50;
+
+/** Default task timeout (10 minutes) */
+const DEFAULT_TASK_TIMEOUT_MS = 10 * 60 * 1000;
+
+/** Minimum tasks in a phase before it gets merged with adjacent */
+const MIN_PHASE_TASKS = 2;
+
+/** TDD phase ordering for grouping */
+const TDD_PHASE_ORDER: Record<TddPhase, number> = {
+  setup: 0,
+  test: 1,
+  impl: 2,
+  verify: 3,
+  review: 4,
+};
+
+// ═══════════════════════════════════════════════════════════════
+// OrchestratorPlanner
+// ═══════════════════════════════════════════════════════════════
+
+export class OrchestratorPlanner {
+  private mux: TerminalMultiplexer;
+  private workingDir: string;
+  private config: OrchestratorConfig;
+  private orchestrator: PlanOrchestrator | null = null;
+
+  constructor(mux: TerminalMultiplexer, workingDir: string, config: OrchestratorConfig) {
+    this.mux = mux;
+    this.workingDir = workingDir;
+    this.config = config;
+  }
+
+  /**
+   * Generate a phased plan from a user goal.
+   *
+   * Uses PlanOrchestrator for AI plan generation, then groups results into phases.
+   */
+  async generatePlan(goal: string, onProgress?: ProgressCallback): Promise<OrchestratorPlan> {
+    const startTime = Date.now();
+
+    // Create a PlanOrchestrator for this plan generation
+    this.orchestrator = new PlanOrchestrator(this.mux, this.workingDir, undefined, {
+      defaultModel: this.config.plannerModel,
+    });
+
+    try {
+      onProgress?.('planning', 'Generating detailed plan...');
+
+      const result: DetailedPlanResult = await this.orchestrator.generateDetailedPlan(goal, onProgress);
+
+      if (!result.success || !result.items || result.items.length === 0) {
+        throw new Error(result.error || 'Plan generation returned no items');
+      }
+
+      // Cap total tasks
+      const items = result.items.slice(0, MAX_TOTAL_TASKS);
+
+      onProgress?.('grouping', 'Organizing plan into phases...');
+
+      // Group items into phases
+      const phases = this.groupIntoPhases(items, goal);
+
+      // Assign team strategies
+      this.assignTeamStrategies(phases);
+
+      // Generate unique completion phrases
+      this.generateCompletionPhrases(phases);
+
+      const plan: OrchestratorPlan = {
+        id: uuidv4(),
+        goal,
+        createdAt: Date.now(),
+        phases,
+        metadata: {
+          totalTasks: phases.reduce((sum, p) => sum + p.tasks.length, 0),
+          estimatedComplexity: this.estimateComplexity(items),
+          modelUsed: this.config.plannerModel,
+          planDurationMs: Date.now() - startTime,
+        },
+      };
+
+      return plan;
+    } finally {
+      this.orchestrator = null;
+    }
+  }
+
+  /** Cancel in-progress plan generation. */
+  async cancel(): Promise<void> {
+    if (this.orchestrator) {
+      await this.orchestrator.cancel();
+      this.orchestrator = null;
+    }
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Phase Grouping
+  // ═══════════════════════════════════════════════════════════════
+
+  /**
+   * Group PlanItems into sequential phases.
+   *
+   * Algorithm:
+   * 1. Build dependency graph and assign IDs to items without them
+   * 2. Topological sort into dependency layers (Kahn's algorithm)
+   * 3. Sub-group within each layer by TDD phase
+   * 4. Merge small phases with their neighbors
+   */
+  private groupIntoPhases(items: PlanItem[], _goal: string): OrchestratorPhase[] {
+    // Ensure all items have IDs
+    const indexedItems = items.map((item, i) => ({
+      ...item,
+      id: item.id || `task-${i}`,
+    }));
+
+    // Build adjacency and in-degree for Kahn's algorithm
+    const idSet = new Set(indexedItems.map((item) => item.id!));
+    const inDegree = new Map<string, number>();
+    const dependents = new Map<string, string[]>(); // id → items that depend on it
+
+    for (const item of indexedItems) {
+      inDegree.set(item.id!, 0);
+      dependents.set(item.id!, []);
+    }
+
+    for (const item of indexedItems) {
+      const deps = (item.dependencies || []).filter((d) => idSet.has(d));
+      inDegree.set(item.id!, deps.length);
+      for (const dep of deps) {
+        dependents.get(dep)!.push(item.id!);
+      }
+    }
+
+    // Kahn's algorithm — produce dependency layers
+    const layers: PlanItem[][] = [];
+    const remaining = new Set(indexedItems.map((item) => item.id!));
+
+    while (remaining.size > 0) {
+      // Find items with no remaining dependencies (in-degree 0)
+      const layer: PlanItem[] = [];
+      for (const id of remaining) {
+        if (inDegree.get(id)! === 0) {
+          layer.push(indexedItems.find((item) => item.id === id)!);
+        }
+      }
+
+      if (layer.length === 0) {
+        // Circular dependency — add all remaining items as a single layer
+        for (const id of remaining) {
+          layer.push(indexedItems.find((item) => item.id === id)!);
+        }
+      }
+
+      layers.push(layer);
+
+      // Remove this layer's items and update in-degrees
+      for (const item of layer) {
+        remaining.delete(item.id!);
+        for (const dep of dependents.get(item.id!) || []) {
+          if (remaining.has(dep)) {
+            inDegree.set(dep, Math.max(0, inDegree.get(dep)! - 1));
+          }
+        }
+      }
+    }
+
+    // Sub-group each layer by TDD phase
+    const rawPhases: PlanItem[][] = [];
+    for (const layer of layers) {
+      const byPhase = new Map<string, PlanItem[]>();
+      for (const item of layer) {
+        const phase = item.tddPhase || 'impl';
+        if (!byPhase.has(phase)) byPhase.set(phase, []);
+        byPhase.get(phase)!.push(item);
+      }
+
+      // Sort sub-groups by TDD phase order
+      const sorted = [...byPhase.entries()].sort(
+        ([a], [b]) => (TDD_PHASE_ORDER[a as TddPhase] ?? 2) - (TDD_PHASE_ORDER[b as TddPhase] ?? 2)
+      );
+
+      for (const [, items] of sorted) {
+        rawPhases.push(items);
+      }
+    }
+
+    // Merge small phases with their previous neighbor
+    const mergedPhases: PlanItem[][] = [];
+    for (const phase of rawPhases) {
+      if (mergedPhases.length > 0 && phase.length < MIN_PHASE_TASKS) {
+        const prev = mergedPhases[mergedPhases.length - 1];
+        if (prev.length < MIN_PHASE_TASKS) {
+          // Merge with previous
+          prev.push(...phase);
+          continue;
+        }
+      }
+      mergedPhases.push([...phase]);
+    }
+
+    // Cap at MAX_PHASES by merging tail phases
+    while (mergedPhases.length > MAX_PHASES) {
+      const last = mergedPhases.pop()!;
+      mergedPhases[mergedPhases.length - 1].push(...last);
+    }
+
+    // Convert to OrchestratorPhase objects
+    return mergedPhases.map((phaseItems, index) => this.createPhase(phaseItems, index));
+  }
+
+  private createPhase(items: PlanItem[], order: number): OrchestratorPhase {
+    // Derive phase name from TDD phases and priorities
+    const tddPhases = [...new Set(items.map((i) => i.tddPhase).filter(Boolean))];
+    const name = this.generatePhaseName(items, tddPhases as TddPhase[], order);
+    const description = items.map((i) => i.content).join('; ');
+
+    const tasks: OrchestratorTask[] = items.map((item, i) => ({
+      id: `phase-${order + 1}-task-${i + 1}`,
+      phaseId: `phase-${order + 1}`,
+      prompt: item.content,
+      status: 'pending' as const,
+      assignedSessionId: null,
+      queueTaskId: null,
+      parallel: items.length > 1, // Tasks within a phase are parallel by default
+      completionPhrase: '', // Assigned later
+      timeoutMs: DEFAULT_TASK_TIMEOUT_MS,
+      startedAt: null,
+      completedAt: null,
+      error: null,
+      retries: 0,
+    }));
+
+    // Extract verification criteria and test commands from items
+    const verificationCriteria = items
+      .map((i) => i.verificationCriteria)
+      .filter((v): v is string => v != null && v.length > 0);
+
+    const testCommands = items.map((i) => i.testCommand).filter((t): t is string => t != null && t.length > 0);
+
+    return {
+      id: `phase-${order + 1}`,
+      name,
+      description,
+      order,
+      status: 'pending' as PhaseStatus,
+      tasks,
+      verificationCriteria,
+      testCommands,
+      maxAttempts: this.config.maxPhaseRetries,
+      attempts: 0,
+      startedAt: null,
+      completedAt: null,
+      durationMs: null,
+      teamStrategy: { type: 'single' }, // Assigned later
+    };
+  }
+
+  private generatePhaseName(items: PlanItem[], tddPhases: TddPhase[], order: number): string {
+    // Try to create a meaningful name based on content
+    const priorities = [...new Set(items.map((i) => i.priority).filter(Boolean))];
+
+    if (tddPhases.length === 1) {
+      const phaseNames: Record<TddPhase, string> = {
+        setup: 'Setup & Configuration',
+        test: 'Test Definition',
+        impl: 'Implementation',
+        verify: 'Verification',
+        review: 'Review & Polish',
+      };
+      return `Phase ${order + 1}: ${phaseNames[tddPhases[0]]}`;
+    }
+
+    if (priorities.includes('P0') && priorities.length === 1) {
+      return `Phase ${order + 1}: Critical Foundation`;
+    }
+
+    return `Phase ${order + 1}: ${items.length > 1 ? 'Parallel Tasks' : items[0].content.slice(0, 50)}`;
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Team Strategy Assignment
+  // ═══════════════════════════════════════════════════════════════
+
+  private assignTeamStrategies(phases: OrchestratorPhase[]): void {
+    for (const phase of phases) {
+      phase.teamStrategy = this.computeTeamStrategy(phase);
+    }
+  }
+
+  private computeTeamStrategy(phase: OrchestratorPhase): TeamStrategy {
+    const taskCount = phase.tasks.length;
+    const parallelTasks = phase.tasks.filter((t) => t.parallel).length;
+
+    // Single task or no parallel potential → single session
+    if (taskCount <= 2 || parallelTasks <= 1) {
+      return { type: 'single' };
+    }
+
+    // If team agents are disabled, use parallel sessions instead
+    if (!this.config.enableTeamAgents) {
+      return {
+        type: 'parallel',
+        maxSessions: Math.min(parallelTasks, this.config.maxParallelSessions),
+      };
+    }
+
+    // 4+ parallel tasks with team agents enabled → team mode
+    if (parallelTasks >= 4) {
+      return {
+        type: 'team',
+        config: {
+          leadPrompt: this.buildTeamLeadPrompt(phase),
+          suggestedTeammates: phase.tasks.slice(0, 4).map((t) => `Specialist for: ${t.prompt.slice(0, 80)}`),
+          maxTeammates: Math.min(parallelTasks, 4),
+        },
+      };
+    }
+
+    // 3 parallel tasks → parallel sessions
+    return {
+      type: 'parallel',
+      maxSessions: Math.min(parallelTasks, this.config.maxParallelSessions),
+    };
+  }
+
+  private buildTeamLeadPrompt(phase: OrchestratorPhase): string {
+    const taskList = phase.tasks.map((t, i) => `${i + 1}. ${t.prompt}`).join('\n');
+
+    return [
+      `You are the team lead for "${phase.name}".`,
+      `Create teammates and delegate the following tasks for parallel execution:`,
+      '',
+      taskList,
+      '',
+      `Each teammate should focus on one task area.`,
+      `When all tasks are complete, verify the results and output: <promise>${phase.id.toUpperCase()}_COMPLETE</promise>`,
+    ].join('\n');
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Completion Phrases
+  // ═══════════════════════════════════════════════════════════════
+
+  private generateCompletionPhrases(phases: OrchestratorPhase[]): void {
+    for (const phase of phases) {
+      for (const task of phase.tasks) {
+        // Generate a unique, deterministic completion phrase per task
+        task.completionPhrase = `ORCH_P${phase.order + 1}_T${phase.tasks.indexOf(task) + 1}`;
+      }
+    }
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Helpers
+  // ═══════════════════════════════════════════════════════════════
+
+  private estimateComplexity(items: PlanItem[]): 'low' | 'medium' | 'high' {
+    const total = items.length;
+    const highComplexity = items.filter((i) => i.complexity === 'high').length;
+    const p0Count = items.filter((i) => i.priority === 'P0').length;
+
+    if (total > 20 || highComplexity > 5 || p0Count > 8) return 'high';
+    if (total > 10 || highComplexity > 2 || p0Count > 4) return 'medium';
+    return 'low';
+  }
+}

src/orchestrator-verifier.ts

@@ -0,0 +1,298 @@
+/**
+ * @fileoverview Orchestrator phase verification.
+ *
+ * Runs verification checks after each phase completes:
+ * - Test commands (shell commands via session)
+ * - AI review (ask Claude to evaluate phase results)
+ *
+ * Three verification modes:
+ * - strict: ALL test commands must pass AND AI review must approve
+ * - moderate: Test commands must pass, AI review is advisory
+ * - lenient: At least one test command passes, AI review skipped
+ *
+ * Key exports:
+ * - `OrchestratorVerifier` class — phase verification engine
+ *
+ * @dependencies types (OrchestratorPhase, VerificationResult, VerificationCheck, OrchestratorConfig)
+ * @consumedby orchestrator-loop
+ *
+ * @module orchestrator-verifier
+ */
+
+import type { Session } from './session.js';
+import {
+  getErrorMessage,
+  type OrchestratorPhase,
+  type OrchestratorConfig,
+  type VerificationResult,
+  type VerificationCheck,
+} from './types.js';
+
+// ═══════════════════════════════════════════════════════════════
+// Constants
+// ═══════════════════════════════════════════════════════════════
+
+/** Timeout for individual test command execution (2 minutes) */
+const TEST_COMMAND_TIMEOUT_MS = 2 * 60 * 1000;
+
+/** Timeout for AI review (3 minutes) */
+const AI_REVIEW_TIMEOUT_MS = 3 * 60 * 1000;
+
+/** Completion phrase for AI verification pass */
+const VERIFY_PASS_PHRASE = 'ORCH_VERIFY_PASS';
+
+/** Completion phrase for AI verification fail */
+const VERIFY_FAIL_PHRASE = 'ORCH_VERIFY_FAIL';
+
+// ═══════════════════════════════════════════════════════════════
+// OrchestratorVerifier
+// ═══════════════════════════════════════════════════════════════
+
+export class OrchestratorVerifier {
+  private config: OrchestratorConfig;
+
+  constructor(config: OrchestratorConfig) {
+    this.config = config;
+  }
+
+  /**
+   * Run all verification checks for a completed phase.
+   *
+   * @param phase - The phase to verify
+   * @param session - Session to use for running commands/reviews
+   * @returns Verification result with pass/fail and suggestions
+   */
+  async verifyPhase(phase: OrchestratorPhase, session: Session): Promise<VerificationResult> {
+    const checks: VerificationCheck[] = [];
+    const mode = this.config.verificationMode;
+
+    // Skip verification entirely in lenient mode with no test commands
+    if (mode === 'lenient' && phase.testCommands.length === 0 && phase.verificationCriteria.length === 0) {
+      return {
+        passed: true,
+        checks: [],
+        summary: 'Verification skipped (lenient mode, no checks defined)',
+        suggestions: [],
+      };
+    }
+
+    // Run test commands if any are defined
+    if (phase.testCommands.length > 0) {
+      const testChecks = await this.runTestCommands(phase.testCommands, session);
+      checks.push(...testChecks);
+    }
+
+    // Run AI review in strict and moderate modes
+    if (mode !== 'lenient' && phase.verificationCriteria.length > 0) {
+      const aiCheck = await this.aiReview(phase, session);
+      checks.push(aiCheck);
+    }
+
+    // Determine pass/fail based on mode
+    const passed = this.evaluateChecks(checks, mode);
+
+    // Generate suggestions for failed checks
+    const suggestions = this.generateSuggestions(checks, phase);
+
+    const passedCount = checks.filter((c) => c.passed).length;
+    const summary =
+      checks.length === 0 ? 'No verification checks defined' : `${passedCount}/${checks.length} checks passed`;
+
+    return { passed, checks, summary, suggestions };
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Test Command Execution
+  // ═══════════════════════════════════════════════════════════════
+
+  private async runTestCommands(commands: string[], session: Session): Promise<VerificationCheck[]> {
+    const checks: VerificationCheck[] = [];
+
+    for (const command of commands) {
+      try {
+        const check = await this.runSingleTestCommand(command, session);
+        checks.push(check);
+      } catch (err) {
+        checks.push({
+          type: 'test_command',
+          description: `Run: ${command}`,
+          passed: false,
+          output: getErrorMessage(err),
+        });
+      }
+    }
+
+    return checks;
+  }
+
+  private async runSingleTestCommand(command: string, session: Session): Promise<VerificationCheck> {
+    // Send the test command to the session and wait for completion
+    // We use a unique marker to detect when the command finishes
+    const marker = `ORCH_TEST_${Date.now()}`;
+    const wrappedCommand = `${command} && echo ${marker}_PASS || echo ${marker}_FAIL`;
+
+    const result = await this.sendAndWaitForMarker(session, wrappedCommand, marker, TEST_COMMAND_TIMEOUT_MS);
+
+    return {
+      type: 'test_command',
+      description: `Run: ${command}`,
+      passed: result.includes(`${marker}_PASS`),
+      output: result.slice(0, 2000), // Truncate output
+    };
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // AI Review
+  // ═══════════════════════════════════════════════════════════════
+
+  private async aiReview(phase: OrchestratorPhase, session: Session): Promise<VerificationCheck> {
+    const prompt = this.buildVerificationPrompt(phase);
+
+    try {
+      const result = await this.sendAndWaitForMarker(
+        session,
+        prompt,
+        VERIFY_PASS_PHRASE,
+        AI_REVIEW_TIMEOUT_MS,
+        VERIFY_FAIL_PHRASE
+      );
+
+      const passed = result.includes(VERIFY_PASS_PHRASE);
+
+      return {
+        type: 'ai_review',
+        description: `AI review of "${phase.name}"`,
+        passed,
+        output: result.slice(0, 3000),
+      };
+    } catch (err) {
+      return {
+        type: 'ai_review',
+        description: `AI review of "${phase.name}"`,
+        passed: false,
+        output: `AI review timed out or failed: ${getErrorMessage(err)}`,
+      };
+    }
+  }
+
+  private buildVerificationPrompt(phase: OrchestratorPhase): string {
+    const criteria = phase.verificationCriteria.map((c, i) => `${i + 1}. ${c}`).join('\n');
+
+    return [
+      `Review the work done in "${phase.name}". Check these criteria:`,
+      '',
+      criteria,
+      '',
+      `If ALL criteria are met, respond with: ${VERIFY_PASS_PHRASE}`,
+      `If ANY criteria fail, respond with: ${VERIFY_FAIL_PHRASE} and explain what failed.`,
+    ].join('\n');
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Evaluation
+  // ═══════════════════════════════════════════════════════════════
+
+  private evaluateChecks(checks: VerificationCheck[], mode: OrchestratorConfig['verificationMode']): boolean {
+    if (checks.length === 0) return true;
+
+    const testChecks = checks.filter((c) => c.type === 'test_command');
+    const aiChecks = checks.filter((c) => c.type === 'ai_review');
+
+    switch (mode) {
+      case 'strict':
+        // ALL checks must pass
+        return checks.every((c) => c.passed);
+
+      case 'moderate':
+        // All test commands must pass; AI review is advisory
+        return testChecks.length === 0 || testChecks.every((c) => c.passed);
+
+      case 'lenient':
+        // At least one test passes (AI review skipped in lenient mode)
+        return testChecks.length === 0 || testChecks.some((c) => c.passed);
+
+      default:
+        return aiChecks.every((c) => c.passed) && testChecks.every((c) => c.passed);
+    }
+  }
+
+  private generateSuggestions(checks: VerificationCheck[], phase: OrchestratorPhase): string[] {
+    const suggestions: string[] = [];
+    const failedChecks = checks.filter((c) => !c.passed);
+
+    if (failedChecks.length === 0) return suggestions;
+
+    for (const check of failedChecks) {
+      if (check.type === 'test_command') {
+        suggestions.push(`Fix failing test: ${check.description}`);
+      } else if (check.type === 'ai_review' && check.output) {
+        // Extract failure reasons from AI review output
+        suggestions.push(`Address AI review feedback for "${phase.name}"`);
+      }
+    }
+
+    return suggestions;
+  }
+
+  // ═══════════════════════════════════════════════════════════════
+  // Session Communication
+  // ═══════════════════════════════════════════════════════════════
+
+  /**
+   * Send a prompt to a session and wait for a marker phrase in the output.
+   *
+   * @param session - Session to send to
+   * @param input - Prompt/command to send
+   * @param marker - Primary marker to watch for
+   * @param timeoutMs - Maximum wait time
+   * @param altMarker - Alternative marker (for pass/fail detection)
+   * @returns Captured output containing the marker
+   */
+  private sendAndWaitForMarker(
+    session: Session,
+    input: string,
+    marker: string,
+    timeoutMs: number,
+    altMarker?: string
+  ): Promise<string> {
+    return new Promise<string>((resolve, reject) => {
+      let output = '';
+      let resolved = false;
+
+      const timer = setTimeout(() => {
+        if (!resolved) {
+          resolved = true;
+          cleanup();
+          reject(new Error(`Timeout waiting for marker "${marker}" after ${timeoutMs}ms`));
+        }
+      }, timeoutMs);
+
+      const handler = (data: string) => {
+        if (resolved) return;
+        output += data;
+
+        if (output.includes(marker) || (altMarker && output.includes(altMarker))) {
+          resolved = true;
+          cleanup();
+          resolve(output);
+        }
+      };
+
+      const cleanup = () => {
+        clearTimeout(timer);
+        session.off('terminal', handler);
+      };
+
+      session.on('terminal', handler);
+
+      // Send the input
+      session.sendInput(input).catch((err) => {
+        if (!resolved) {
+          resolved = true;
+          cleanup();
+          reject(err);
+        }
+      });
+    });
+  }
+}

src/plan-orchestrator.ts

@@ -20,36 +20,10 @@ import type { TerminalMultiplexer } from './mux-interface.js';
 import { existsSync, mkdirSync, writeFileSync } from 'node:fs';
 import { join } from 'node:path';
 import { RESEARCH_AGENT_PROMPT, PLANNER_PROMPT } from './prompts/index.js';
-import { PlanTaskStatus, TddPhase } from './types.js';
+import { getErrorMessage, type PlanItem } from './types.js';
 
-// ============================================================================
-// Types
-// ============================================================================
-
-/** Development phase in TDD cycle (alias for TddPhase) */
-export type PlanPhase = TddPhase;
-
-/**
- * Plan item with TDD structure.
- */
-export interface PlanItem {
-  id?: string;
-  content: string;
-  priority: 'P0' | 'P1' | 'P2' | null;
-  source?: string;
-  rationale?: string;
-  verificationCriteria?: string;
-  testCommand?: string;
-  dependencies?: string[];
-  status?: PlanTaskStatus;
-  attempts?: number;
-  lastError?: string;
-  completedAt?: number;
-  complexity?: 'low' | 'medium' | 'high';
-  tddPhase?: PlanPhase;
-  pairedWith?: string;
-  reviewChecklist?: string[];
-}
+// Re-export for backward compatibility
+export type { PlanItem };
 
 export interface ResearchResult {
   success: boolean;
@@ -257,6 +231,49 @@ export class PlanOrchestrator {
     return md;
   }
 
+  private _extractJsonFromResponse(response: string): string | null {
+    let jsonMatch = response.match(/```(?:json)?\s*(\{[\s\S]*?\})\s*```/);
+    if (jsonMatch) {
+      jsonMatch = [jsonMatch[1]]; // Use captured group (inside code block)
+    } else {
+      jsonMatch = response.match(/\{[\s\S]*\}/);
+    }
+    return jsonMatch ? jsonMatch[0] : null;
+  }
+
+  private _emitAgentFailure(
+    onSubagent: SubagentCallback | undefined,
+    agentId: string,
+    agentType: 'research' | 'planner',
+    model: string,
+    error: string,
+    durationMs: number
+  ): void {
+    onSubagent?.({
+      type: 'failed',
+      agentId,
+      agentType,
+      model,
+      status: 'failed',
+      error,
+      durationMs,
+    });
+  }
+
+  private _formatResearchSection(
+    parts: string[],
+    title: string,
+    items: unknown[],
+    formatter: (item: unknown) => string[]
+  ): void {
+    if (items.length === 0) return;
+    parts.push(title);
+    for (const item of items.slice(0, 5)) {
+      parts.push(...formatter(item));
+    }
+    parts.push('');
+  }
+
   async cancel(): Promise<void> {
     this.cancelled = true;
     // Stop all running sessions and await cleanup to prevent PTY process leaks
@@ -338,7 +355,7 @@ export class PlanOrchestrator {
     } catch (err) {
       return {
         success: false,
-        error: err instanceof Error ? err.message : String(err),
+        error: getErrorMessage(err),
       };
     }
   }
@@ -348,32 +365,23 @@ export class PlanOrchestrator {
 
     const parts: string[] = ['## Research Context\n'];
 
-    if (research.findings.externalResources.length > 0) {
-      parts.push('### External Resources');
-      for (const r of research.findings.externalResources.slice(0, 5)) {
-        parts.push(`- ${r.title}${r.url ? ` (${r.url})` : ''}`);
-        if (r.keyInsights.length > 0) {
-          parts.push(`  Key insights: ${r.keyInsights.slice(0, 3).join(', ')}`);
-        }
+    this._formatResearchSection(parts, '### External Resources', research.findings.externalResources, (item) => {
+      const r = item as ResearchResult['findings']['externalResources'][number];
+      const lines = [`- ${r.title}${r.url ? ` (${r.url})` : ''}`];
+      if (r.keyInsights.length > 0) {
+        lines.push(`  Key insights: ${r.keyInsights.slice(0, 3).join(', ')}`);
       }
-      parts.push('');
-    }
+      return lines;
+    });
 
-    if (research.findings.codebasePatterns.length > 0) {
-      parts.push('### Existing Codebase Patterns');
-      for (const p of research.findings.codebasePatterns.slice(0, 5)) {
-        parts.push(`- ${p.pattern} at ${p.location}`);
-      }
-      parts.push('');
-    }
+    this._formatResearchSection(parts, '### Existing Codebase Patterns', research.findings.codebasePatterns, (item) => {
+      const p = item as ResearchResult['findings']['codebasePatterns'][number];
+      return [`- ${p.pattern} at ${p.location}`];
+    });
 
-    if (research.findings.technicalRecommendations.length > 0) {
-      parts.push('### Recommendations');
-      for (const r of research.findings.technicalRecommendations.slice(0, 5)) {
-        parts.push(`- ${r}`);
-      }
-      parts.push('');
-    }
+    this._formatResearchSection(parts, '### Recommendations', research.findings.technicalRecommendations, (item) => [
+      `- ${item as string}`,
+    ]);
 
     return parts.join('\n');
   }
@@ -440,18 +448,20 @@ export class PlanOrchestrator {
 
       const durationMs = Date.now() - startTime;
 
-      // Extract JSON from response
-      const jsonMatch = response.match(/\{[\s\S]*\}/);
-      if (!jsonMatch) {
-        onSubagent?.({
-          type: 'failed',
-          agentId,
-          agentType: 'research',
-          model: this.researchModel,
-          status: 'failed',
-          error: 'No JSON found',
-          durationMs,
-        });
+      console.log(
+        `[PlanOrchestrator] Research response length: ${response.length}, first 500 chars:`,
+        response.substring(0, 500)
+      );
+
+      // Extract JSON from response — try multiple strategies
+      const jsonStr = this._extractJsonFromResponse(response);
+
+      if (!jsonStr) {
+        console.error(
+          `[PlanOrchestrator] No JSON found in research response. Full response:`,
+          response.substring(0, 2000)
+        );
+        this._emitAgentFailure(onSubagent, agentId, 'research', this.researchModel, 'No JSON found', durationMs);
         return {
           success: false,
           findings: {
@@ -467,17 +477,9 @@ export class PlanOrchestrator {
         };
       }
 
-      const parsed = tryParseJSON(jsonMatch[0]);
+      const parsed = tryParseJSON(jsonStr);
       if (!parsed.success) {
-        onSubagent?.({
-          type: 'failed',
-          agentId,
-          agentType: 'research',
-          model: this.researchModel,
-          status: 'failed',
-          error: parsed.error,
-          durationMs,
-        });
+        this._emitAgentFailure(onSubagent, agentId, 'research', this.researchModel, parsed.error!, durationMs);
         return {
           success: false,
           findings: {
@@ -521,16 +523,8 @@ export class PlanOrchestrator {
       return result;
     } catch (err) {
       const durationMs = Date.now() - startTime;
-      const error = err instanceof Error ? err.message : String(err);
-      onSubagent?.({
-        type: 'failed',
-        agentId,
-        agentType: 'research',
-        model: this.researchModel,
-        status: 'failed',
-        error,
-        durationMs,
-      });
+      const error = getErrorMessage(err);
+      this._emitAgentFailure(onSubagent, agentId, 'research', this.researchModel, error, durationMs);
       return {
         success: false,
         findings: {
@@ -547,7 +541,7 @@ export class PlanOrchestrator {
     } finally {
       // Always clean up session and progress interval — centralizing here
       // prevents the race where cancel() and catch both try to manage the set
-      await session.stop().catch(() => {});
+      await session.stop().catch(() => {}); // Ignore - session cleanup is best-effort in finally block
       this.runningSessions.delete(session);
       clearInterval(progressInterval);
     }
@@ -613,32 +607,26 @@ export class PlanOrchestrator {
 
       const durationMs = Date.now() - startTime;
 
-      // Extract JSON from response
-      const jsonMatch = response.match(/\{[\s\S]*\}/);
-      if (!jsonMatch) {
-        onSubagent?.({
-          type: 'failed',
-          agentId,
-          agentType: 'planner',
-          model: this.plannerModel,
-          status: 'failed',
-          error: 'No JSON found',
-          durationMs,
-        });
+      console.log(
+        `[PlanOrchestrator] Planner response length: ${response.length}, first 500 chars:`,
+        response.substring(0, 500)
+      );
+
+      // Extract JSON from response — try multiple strategies
+      const jsonStr = this._extractJsonFromResponse(response);
+
+      if (!jsonStr) {
+        console.error(
+          `[PlanOrchestrator] No JSON found in planner response. Full response:`,
+          response.substring(0, 2000)
+        );
+        this._emitAgentFailure(onSubagent, agentId, 'planner', this.plannerModel, 'No JSON found', durationMs);
         return { success: false, error: 'No JSON in response' };
       }
 
-      const parsed = tryParseJSON(jsonMatch[0]);
+      const parsed = tryParseJSON(jsonStr);
       if (!parsed.success) {
-        onSubagent?.({
-          type: 'failed',
-          agentId,
-          agentType: 'planner',
-          model: this.plannerModel,
-          status: 'failed',
-          error: parsed.error,
-          durationMs,
-        });
+        this._emitAgentFailure(onSubagent, agentId, 'planner', this.plannerModel, parsed.error!, durationMs);
         return { success: false, error: parsed.error };
       }
 
@@ -663,21 +651,13 @@ export class PlanOrchestrator {
       return { success: true, items, gaps, warnings };
     } catch (err) {
       const durationMs = Date.now() - startTime;
-      const error = err instanceof Error ? err.message : String(err);
-      onSubagent?.({
-        type: 'failed',
-        agentId,
-        agentType: 'planner',
-        model: this.plannerModel,
-        status: 'failed',
-        error,
-        durationMs,
-      });
+      const error = getErrorMessage(err);
+      this._emitAgentFailure(onSubagent, agentId, 'planner', this.plannerModel, error, durationMs);
       return { success: false, error };
     } finally {
       // Always clean up session and progress interval — centralizing here
       // prevents the race where cancel() and catch both try to manage the set
-      await session.stop().catch(() => {});
+      await session.stop().catch(() => {}); // Ignore - session cleanup is best-effort in finally block
       this.runningSessions.delete(session);
       clearInterval(progressInterval);
     }

src/prompts/index.ts

@@ -7,3 +7,10 @@
 
 export { RESEARCH_AGENT_PROMPT } from './research-agent.js';
 export { PLANNER_PROMPT } from './planner.js';
+export {
+  PHASE_EXECUTION_PROMPT,
+  TEAM_LEAD_PROMPT,
+  VERIFICATION_PROMPT,
+  REPLAN_PROMPT,
+  SINGLE_TASK_PROMPT,
+} from './orchestrator.js';

src/prompts/orchestrator.ts

@@ -0,0 +1,116 @@
+/**
+ * @fileoverview Orchestrator Loop prompt templates.
+ *
+ * Templates for phase execution, team delegation, verification, and replanning.
+ * Placeholders use {VARIABLE} syntax and are replaced at runtime.
+ *
+ * @module prompts/orchestrator
+ */
+
+/**
+ * Phase execution prompt — tells Claude what to accomplish in this phase.
+ *
+ * Placeholders:
+ * - {PHASE_NUMBER}: Phase index (1-based)
+ * - {PHASE_NAME}: Human-readable phase name
+ * - {GOAL}: Original user goal
+ * - {COMPLETED_PHASES}: Summary of previously completed phases
+ * - {TASK_LIST}: Numbered task list for this phase
+ * - {VERIFICATION_CRITERIA}: What will be checked after this phase
+ * - {COMPLETION_PHRASE}: The phrase to output when done
+ */
+export const PHASE_EXECUTION_PROMPT = `You are executing {PHASE_NAME} of a larger project.
+
+OVERALL GOAL: {GOAL}
+
+COMPLETED SO FAR:
+{COMPLETED_PHASES}
+
+YOUR TASKS FOR THIS PHASE:
+{TASK_LIST}
+
+Complete each task thoroughly. Run tests after each change to catch issues early.
+
+VERIFICATION (will be checked after you finish):
+{VERIFICATION_CRITERIA}
+
+When ALL tasks in this phase are complete and verified, output: <promise>{COMPLETION_PHRASE}</promise>`;
+
+/**
+ * Team lead delegation prompt — instructs a lead to coordinate teammates.
+ *
+ * Placeholders:
+ * - {PHASE_NAME}: Phase name
+ * - {TASK_LIST}: Numbered task list
+ * - {TEAMMATE_HINTS}: Suggested teammate specializations
+ * - {COMPLETION_PHRASE}: Phrase for when all work is done
+ */
+export const TEAM_LEAD_PROMPT = `You are the team lead for {PHASE_NAME}.
+
+Create teammates and delegate the following tasks for parallel execution:
+
+{TASK_LIST}
+
+Suggested teammate roles:
+{TEAMMATE_HINTS}
+
+Each teammate should focus on their assigned task area. Monitor their progress.
+When ALL tasks are complete and you've verified the results, output: <promise>{COMPLETION_PHRASE}</promise>`;
+
+/**
+ * Verification prompt — asks Claude to verify phase completion.
+ *
+ * Placeholders:
+ * - {PHASE_NAME}: Phase name
+ * - {CRITERIA}: Numbered verification criteria
+ * - {PASS_PHRASE}: Phrase to output on success
+ * - {FAIL_PHRASE}: Phrase to output on failure
+ */
+export const VERIFICATION_PROMPT = `Review the work done in "{PHASE_NAME}". Check these criteria:
+
+{CRITERIA}
+
+If ALL criteria are met, respond with: {PASS_PHRASE}
+If ANY criteria fail, respond with: {FAIL_PHRASE} and explain what failed.`;
+
+/**
+ * Replan prompt — gives failure context and asks for recovery.
+ *
+ * Placeholders:
+ * - {PHASE_NAME}: Phase name
+ * - {ATTEMPT_NUMBER}: Current retry attempt
+ * - {MAX_ATTEMPTS}: Maximum attempts allowed
+ * - {FAILURE_SUMMARY}: What went wrong
+ * - {SUGGESTIONS}: Recovery suggestions from verification
+ * - {ORIGINAL_TASKS}: The original task list
+ * - {COMPLETION_PHRASE}: Phrase for when recovery is done
+ */
+export const REPLAN_PROMPT = `Phase "{PHASE_NAME}" verification failed (attempt {ATTEMPT_NUMBER}/{MAX_ATTEMPTS}).
+
+WHAT WENT WRONG:
+{FAILURE_SUMMARY}
+
+SUGGESTIONS:
+{SUGGESTIONS}
+
+ORIGINAL TASKS:
+{ORIGINAL_TASKS}
+
+Fix the issues identified above. Focus on making the verification criteria pass.
+When the fixes are complete, output: <promise>{COMPLETION_PHRASE}</promise>`;
+
+/**
+ * Single-task execution prompt — for phases with a single task.
+ *
+ * Placeholders:
+ * - {TASK}: The task description
+ * - {GOAL}: Original user goal
+ * - {CONTEXT}: Any relevant context
+ * - {COMPLETION_PHRASE}: Phrase for when done
+ */
+export const SINGLE_TASK_PROMPT = `{TASK}
+
+Context: This is part of a larger project — {GOAL}
+{CONTEXT}
+
+When done, output: <promise>{COMPLETION_PHRASE}</promise>`;

src/prompts/planner.ts

@@ -1,14 +1,13 @@
 /**
- * Planner Prompt - Single agent for TDD plan generation
- *
- * Combines what was previously 5 separate agents:
- * - Requirements Analyst (redundant)
- * - Architecture Planner (redundant)
- * - Testing Specialist (kept - TDD focus)
- * - Risk Analyst (redundant)
- * - Verification Expert (kept - structure)
+ * @fileoverview Planner Prompt — single TDD plan generator combining
+ * requirements analysis, architecture, testing, risk, and verification
+ * into one agent (previously 5 separate agents).
  *
  * Placeholders: {TASK}, {RESEARCH_CONTEXT}
+ *
+ * @dependencies none (pure template)
+ * @consumedby prompts/index (re-export), plan-orchestrator
+ * @module prompts/planner
  */
 
 export const PLANNER_PROMPT = `You are a TDD Plan Generator. Create a complete implementation plan with test-first approach.

src/prompts/research-agent.ts

@@ -1,10 +1,12 @@
 /**
- * Research Agent Prompt
- *
- * Gathers external resources, codebase patterns, and technical context
- * before other agents analyze the task.
+ * @fileoverview Research Agent Prompt — gathers codebase patterns, external
+ * resources, and technical context before the planner analyzes the task.
  *
  * Placeholders: {TASK}, {WORKING_DIR}
+ *
+ * @dependencies none (pure template)
+ * @consumedby prompts/index (re-export), plan-orchestrator
+ * @module prompts/research-agent
  */
 
 export const RESEARCH_AGENT_PROMPT = `You are a Research Specialist preparing context for an implementation task. Your job is to gather all relevant information that will help the development team succeed.

src/push-store.ts

@@ -11,6 +11,7 @@ import { join } from 'node:path';
 import { homedir } from 'node:os';
 import webpush from 'web-push';
 import type { VapidKeys, PushSubscriptionRecord } from './types.js';
+import { Debouncer } from './utils/index.js';
 
 const DATA_DIR = join(homedir(), '.codeman');
 const KEYS_FILE = join(DATA_DIR, 'push-keys.json');
@@ -20,7 +21,7 @@ const SAVE_DEBOUNCE_MS = 500;
 export class PushSubscriptionStore {
   private vapidKeys: VapidKeys | null = null;
   private subscriptions: Map<string, PushSubscriptionRecord> = new Map();
-  private saveTimer: NodeJS.Timeout | null = null;
+  private saveDeb = new Debouncer(SAVE_DEBOUNCE_MS);
   private _disposed = false;
 
   constructor() {
@@ -149,10 +150,7 @@ export class PushSubscriptionStore {
   /** Schedule a debounced save */
   private scheduleSave(): void {
     if (this._disposed) return;
-    if (this.saveTimer) clearTimeout(this.saveTimer);
-    this.saveTimer = setTimeout(() => {
-      this.flushSave();
-    }, SAVE_DEBOUNCE_MS);
+    this.saveDeb.schedule(() => this.flushSave());
   }
 
   /** Immediately persist subscriptions to disk */
@@ -169,11 +167,6 @@ export class PushSubscriptionStore {
   dispose(): void {
     if (this._disposed) return;
     this._disposed = true;
-    if (this.saveTimer) {
-      clearTimeout(this.saveTimer);
-      this.saveTimer = null;
-    }
-    // Final flush
-    this.flushSave();
+    this.saveDeb.flush(() => this.flushSave());
   }
 }

src/ralph-config.ts

@@ -9,6 +9,7 @@
 
 import { existsSync, readFileSync } from 'node:fs';
 import { join } from 'node:path';
+import { execPattern } from './utils/index.js';
 
 // Pattern to extract completion phrase from CLAUDE.md
 // Matches <promise>PHRASE</promise> with optional whitespace
@@ -83,9 +84,7 @@ export function parseRalphLoopConfigFromContent(content: string): RalphLoopConfi
   };
 
   // Parse each YAML line
-  let match;
-  YAML_LINE_PATTERN.lastIndex = 0;
-  while ((match = YAML_LINE_PATTERN.exec(yaml)) !== null) {
+  execPattern(YAML_LINE_PATTERN, yaml, (match) => {
     const key = match[1].toLowerCase();
     const value = match[2].trim();
 
@@ -103,7 +102,7 @@ export function parseRalphLoopConfigFromContent(content: string): RalphLoopConfi
         config.completionPromise = value.toUpperCase();
         break;
     }
-  }
+  });
 
   return config;
 }

src/ralph-fix-plan-watcher.ts

@@ -0,0 +1,366 @@
+/**
+ * @fileoverview RalphFixPlanWatcher - Watches @fix_plan.md for changes
+ *
+ * Monitors the @fix_plan.md file in the session's working directory
+ * for changes, parsing todo items from the markdown format.
+ *
+ * Extracted from ralph-tracker.ts as part of domain splitting.
+ *
+ * @module ralph-fix-plan-watcher
+ */
+
+import { EventEmitter } from 'node:events';
+import { readFile } from 'node:fs/promises';
+import { existsSync, FSWatcher, watch as fsWatch } from 'node:fs';
+import { join } from 'node:path';
+import type { RalphTodoStatus, RalphTodoPriority, RalphTodoItem } from './types.js';
+
+// ========== @fix_plan.md Generation & Import Utility Functions ==========
+
+/**
+ * Generate @fix_plan.md content from todo items.
+ * Groups todos by priority and status.
+ *
+ * @param todos - Array of todo items
+ * @returns Markdown content for @fix_plan.md
+ */
+export function generateFixPlanMarkdown(todos: RalphTodoItem[]): string {
+  const lines: string[] = ['# Fix Plan', ''];
+
+  // Group by priority
+  const p0: RalphTodoItem[] = [];
+  const p1: RalphTodoItem[] = [];
+  const p2: RalphTodoItem[] = [];
+  const noPriority: RalphTodoItem[] = [];
+  const completed: RalphTodoItem[] = [];
+
+  for (const todo of todos) {
+    if (todo.status === 'completed') {
+      completed.push(todo);
+    } else if (todo.priority === 'P0') {
+      p0.push(todo);
+    } else if (todo.priority === 'P1') {
+      p1.push(todo);
+    } else if (todo.priority === 'P2') {
+      p2.push(todo);
+    } else {
+      noPriority.push(todo);
+    }
+  }
+
+  // High Priority (P0)
+  if (p0.length > 0) {
+    lines.push('## High Priority (P0)');
+    for (const todo of p0) {
+      const checkbox = todo.status === 'in_progress' ? '[-]' : '[ ]';
+      lines.push(`- ${checkbox} ${todo.content}`);
+    }
+    lines.push('');
+  }
+
+  // Standard (P1)
+  if (p1.length > 0) {
+    lines.push('## Standard (P1)');
+    for (const todo of p1) {
+      const checkbox = todo.status === 'in_progress' ? '[-]' : '[ ]';
+      lines.push(`- ${checkbox} ${todo.content}`);
+    }
+    lines.push('');
+  }
+
+  // Nice to Have (P2)
+  if (p2.length > 0) {
+    lines.push('## Nice to Have (P2)');
+    for (const todo of p2) {
+      const checkbox = todo.status === 'in_progress' ? '[-]' : '[ ]';
+      lines.push(`- ${checkbox} ${todo.content}`);
+    }
+    lines.push('');
+  }
+
+  // Tasks (no priority)
+  if (noPriority.length > 0) {
+    lines.push('## Tasks');
+    for (const todo of noPriority) {
+      const checkbox = todo.status === 'in_progress' ? '[-]' : '[ ]';
+      lines.push(`- ${checkbox} ${todo.content}`);
+    }
+    lines.push('');
+  }
+
+  // Completed
+  if (completed.length > 0) {
+    lines.push('## Completed');
+    for (const todo of completed) {
+      lines.push(`- [x] ${todo.content}`);
+    }
+    lines.push('');
+  }
+
+  return lines.join('\n');
+}
+
+/**
+ * Parse @fix_plan.md content and return parsed todo items.
+ *
+ * @param content - Markdown content from @fix_plan.md
+ * @param parsePriority - Function to parse priority from content text
+ * @param generateTodoId - Function to generate stable todo ID from content
+ * @returns Array of parsed todo items
+ */
+export function importFixPlanMarkdown(
+  content: string,
+  parsePriority: (content: string) => RalphTodoPriority,
+  generateTodoId: (content: string) => string
+): RalphTodoItem[] {
+  const lines = content.split('\n');
+  const newTodos: RalphTodoItem[] = [];
+  let currentPriority: RalphTodoPriority = null;
+
+  // Patterns for section headers
+  const p0HeaderPattern = /^##\s*(High Priority|Critical|P0)/i;
+  const p1HeaderPattern = /^##\s*(Standard|P1|Medium Priority)/i;
+  const p2HeaderPattern = /^##\s*(Nice to Have|P2|Low Priority)/i;
+  const completedHeaderPattern = /^##\s*Completed/i;
+  const tasksHeaderPattern = /^##\s*Tasks/i;
+
+  // Pattern for todo items
+  const todoPattern = /^-\s*\[([ x-])\]\s*(.+)$/;
+
+  let inCompletedSection = false;
+
+  for (const line of lines) {
+    const trimmed = line.trim();
+
+    // Check for section headers
+    if (p0HeaderPattern.test(trimmed)) {
+      currentPriority = 'P0';
+      inCompletedSection = false;
+      continue;
+    }
+    if (p1HeaderPattern.test(trimmed)) {
+      currentPriority = 'P1';
+      inCompletedSection = false;
+      continue;
+    }
+    if (p2HeaderPattern.test(trimmed)) {
+      currentPriority = 'P2';
+      inCompletedSection = false;
+      continue;
+    }
+    if (completedHeaderPattern.test(trimmed)) {
+      inCompletedSection = true;
+      continue;
+    }
+    if (tasksHeaderPattern.test(trimmed)) {
+      currentPriority = null;
+      inCompletedSection = false;
+      continue;
+    }
+
+    // Parse todo item
+    const match = trimmed.match(todoPattern);
+    if (match) {
+      const [, checkboxState, todoContent] = match;
+      let status: RalphTodoStatus;
+
+      if (inCompletedSection || checkboxState === 'x' || checkboxState === 'X') {
+        status = 'completed';
+      } else if (checkboxState === '-') {
+        status = 'in_progress';
+      } else {
+        status = 'pending';
+      }
+
+      // Parse priority from content if not in a priority section
+      const parsedPriority = inCompletedSection ? null : currentPriority || parsePriority(todoContent);
+
+      const id = generateTodoId(todoContent);
+      newTodos.push({
+        id,
+        content: todoContent.trim(),
+        status,
+        detectedAt: Date.now(),
+        priority: parsedPriority,
+      });
+    }
+  }
+
+  return newTodos;
+}
+
+/**
+ * RalphFixPlanWatcher - Watches @fix_plan.md for changes.
+ *
+ * Events emitted:
+ * - `todosLoaded` - Emits parsed todo items when @fix_plan.md is loaded/changed
+ * - `enabled` - Emits when tracker should be auto-enabled (todos loaded from file)
+ */
+export class RalphFixPlanWatcher extends EventEmitter {
+  /** Working directory for @fix_plan.md watching */
+  private _workingDir: string | null = null;
+
+  /** Path to the @fix_plan.md file being watched */
+  private _fixPlanPath: string | null = null;
+
+  /** File watcher for @fix_plan.md */
+  private _fixPlanWatcher: FSWatcher | null = null;
+
+  /** Error handler for FSWatcher (stored for cleanup to prevent memory leak) */
+  private _fixPlanWatcherErrorHandler: ((err: Error) => void) | null = null;
+
+  /** Debounce timer for file change events */
+  private _fixPlanReloadTimer: NodeJS.Timeout | null = null;
+
+  /** Priority parser injected from parent (for importFixPlanMarkdown) */
+  private _parsePriority: (content: string) => RalphTodoPriority;
+
+  /** Todo ID generator injected from parent */
+  private _generateTodoId: (content: string) => string;
+
+  constructor(parsePriority: (content: string) => RalphTodoPriority, generateTodoId: (content: string) => string) {
+    super();
+    this._parsePriority = parsePriority;
+    this._generateTodoId = generateTodoId;
+  }
+
+  /**
+   * When @fix_plan.md is active, treat it as the source of truth for todo status.
+   * This prevents output-based detection from overriding file-based status.
+   */
+  get isFileAuthoritative(): boolean {
+    return this._fixPlanPath !== null;
+  }
+
+  /**
+   * Set the working directory and start watching @fix_plan.md.
+   * Automatically loads existing @fix_plan.md if present.
+   * @param workingDir - The session's working directory
+   */
+  setWorkingDir(workingDir: string): void {
+    this._workingDir = workingDir;
+    this._fixPlanPath = join(workingDir, '@fix_plan.md');
+
+    // Try to load existing @fix_plan.md
+    this.loadFixPlanFromDisk();
+
+    // Start watching for changes
+    this.startWatchingFixPlan();
+  }
+
+  /**
+   * Load @fix_plan.md from disk if it exists.
+   * Called on initialization and when file changes are detected.
+   */
+  async loadFixPlanFromDisk(): Promise<number> {
+    if (!this._fixPlanPath) return 0;
+
+    try {
+      if (!existsSync(this._fixPlanPath)) {
+        return 0;
+      }
+
+      const content = await readFile(this._fixPlanPath, 'utf-8');
+      const todos = importFixPlanMarkdown(content, this._parsePriority, this._generateTodoId);
+
+      if (todos.length > 0) {
+        this.emit('todosLoaded', todos);
+        console.log(`[RalphFixPlanWatcher] Loaded ${todos.length} todos from @fix_plan.md`);
+      }
+
+      return todos.length;
+    } catch (err) {
+      // File doesn't exist or can't be read - that's OK
+      console.log(`[RalphFixPlanWatcher] Could not load @fix_plan.md: ${err}`);
+      return 0;
+    }
+  }
+
+  /**
+   * Start watching @fix_plan.md for changes.
+   * Reloads todos when the file is modified.
+   */
+  private startWatchingFixPlan(): void {
+    if (!this._fixPlanPath || !this._workingDir) return;
+
+    // Stop existing watcher if any
+    this.stopWatchingFixPlan();
+
+    try {
+      // Only watch if the file exists
+      if (!existsSync(this._fixPlanPath)) {
+        // Watch the directory instead for file creation
+        this._fixPlanWatcher = fsWatch(this._workingDir, (_eventType, filename) => {
+          if (filename === '@fix_plan.md') {
+            this.handleFixPlanChange();
+          }
+        });
+      } else {
+        // Watch the file directly
+        this._fixPlanWatcher = fsWatch(this._fixPlanPath, () => {
+          this.handleFixPlanChange();
+        });
+      }
+      // Add error handler to prevent unhandled errors and clean up on failure
+      // Store handler reference for proper cleanup in stopWatchingFixPlan()
+      if (this._fixPlanWatcher) {
+        this._fixPlanWatcherErrorHandler = (err: Error) => {
+          console.log(`[RalphFixPlanWatcher] FSWatcher error for @fix_plan.md: ${err.message}`);
+          this.stopWatchingFixPlan();
+        };
+        this._fixPlanWatcher.on('error', this._fixPlanWatcherErrorHandler);
+      }
+    } catch (err) {
+      console.log(`[RalphFixPlanWatcher] Could not watch @fix_plan.md: ${err}`);
+    }
+  }
+
+  /**
+   * Handle @fix_plan.md file change with debouncing.
+   */
+  private handleFixPlanChange(): void {
+    // Debounce rapid changes (e.g., multiple writes)
+    if (this._fixPlanReloadTimer) {
+      clearTimeout(this._fixPlanReloadTimer);
+    }
+
+    this._fixPlanReloadTimer = setTimeout(() => {
+      this._fixPlanReloadTimer = null;
+      this.loadFixPlanFromDisk();
+    }, 500); // 500ms debounce
+  }
+
+  /**
+   * Stop watching @fix_plan.md.
+   */
+  stopWatchingFixPlan(): void {
+    if (this._fixPlanWatcher) {
+      // Remove error handler before closing to prevent memory leak
+      if (this._fixPlanWatcherErrorHandler) {
+        this._fixPlanWatcher.off('error', this._fixPlanWatcherErrorHandler);
+        this._fixPlanWatcherErrorHandler = null;
+      }
+      this._fixPlanWatcher.close();
+      this._fixPlanWatcher = null;
+    }
+    if (this._fixPlanReloadTimer) {
+      clearTimeout(this._fixPlanReloadTimer);
+      this._fixPlanReloadTimer = null;
+    }
+  }
+
+  /**
+   * Stop watching and clean up all resources.
+   */
+  stop(): void {
+    this.stopWatchingFixPlan();
+  }
+
+  /**
+   * Clean up all resources.
+   */
+  destroy(): void {
+    this.stop();
+    this.removeAllListeners();
+  }
+}

src/ralph-loop.ts

@@ -1,14 +1,24 @@
 /**
- * @fileoverview Ralph Loop - Autonomous task execution engine
+ * @fileoverview Ralph Loop - Autonomous task execution engine.
  *
- * The Ralph Loop orchestrates autonomous Claude sessions by:
+ * Orchestrates autonomous Claude sessions by:
  * - Polling for available tasks from the task queue
  * - Assigning tasks to idle sessions
  * - Monitoring completion and handling failures
  * - Auto-generating follow-up tasks when min duration not reached
  *
- * Named after Ralph Wiggum's persistence ("I'm in danger!"),
- * this loop keeps Claude working until all tasks are done.
+ * Key exports:
+ * - `RalphLoop` class — the loop engine, extends EventEmitter
+ * - `RalphLoopEvents` interface — typed event map
+ * - `RalphLoopOptions` interface — configuration options
+ *
+ * Lifecycle: `start()` → poll loop → `stop()` (when all tasks done + min duration met)
+ *
+ * @dependencies session-manager (session lifecycle), task-queue (task FIFO),
+ *   state-store (persistence), session (PTY execution), task (task model)
+ * @consumedby web/server (ralph routes, SSE)
+ * @emits started, stopped, taskAssigned, taskCompleted, taskFailed, error
+ * @persistence Ralph loop state saved to `~/.codeman/state.json` (ralphLoop key)
  *
  * @module ralph-loop
  */