chore: version packages

The reliable-delivery layer marks every keystroke as briefly pending until its
ACK lands a few ms later, which made the connection indicator flash
"Sending 1B…" on every character during normal typing. Hide the indicator
entirely while the connection is healthy (connected/connecting) — it now only
appears for an actual problem (reconnecting/offline), where the queued-byte
count reassures the user their input is safely buffered.

Verified in a real browser: hidden throughout connected typing, shows
"Offline (NB queued)" when offline, hides again after reconnect+delivery.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -1,5 +1,53 @@
 # aicodeman
 
+## 1.1.17
+
+### Patch Changes
+
+- Fix the connection indicator flashing "Sending 1B…" on every keystroke. The reliable input-delivery layer (1.1.16) marks each keystroke as briefly pending until its ACK arrives a few milliseconds later, which made the indicator flash on every character while typing on a healthy connection. The indicator is now hidden whenever the connection is healthy and only appears for an actual problem (reconnecting/offline), where it still shows the queued byte count so you know buffered input will be sent.
+
+## 1.1.16
+
+### Patch Changes
+
+- Mobile image uploads, reliable input delivery, and gesture window dragging.
+
+  **Mobile image uploads (camera-roll picker / drag-drop / paste).** The "🖼 Image" button now handles real photo batches: up to 20 images per batch uploaded with bounded concurrency and a live "Uploading N/M…" progress toast (with a summary of successes, failures, and whether the 20-cap trimmed the selection). The per-file limit is raised from 10MB to 50MB (`MAX_PASTE_IMAGE_BYTES`, env-overridable via `CODEMAN_MAX_PASTE_IMAGE_BYTES`) so full-resolution phone photos and large screenshots are accepted. Very large images are downscaled to ≤4096px on the longest edge before upload, fixing iOS Safari's ~16.7M-px `<canvas>` limit that previously made huge photos fail to re-encode. Also fixes a latent concurrency bug the batch path exposed where the first parallel uploads to a session raced on creating `.claude-images/` and failed with EEXIST.
+
+  **Reliable, exactly-once input delivery.** A "sent" prompt could be silently lost on a flaky connection (e.g. a train): a half-open WebSocket accepts `ws.send()` without error while discarding the frame, and nothing was queued or resent. Input is now recorded durably (localStorage) with a stable clientId + monotonic per-session sequence before delivery, and only dropped once the server ACKs it — delivered over the WebSocket (acked via `{t:'ia',seq}`) or, when the socket is down, over POST in order. A 2s sweep force-reconnects a half-open socket; pending input survives reconnects and page reloads. The server applies each `(clientId, seq)` at most once (`Session.shouldApplyInput`), so an at-least-once resend can never type the prompt twice. Untagged input (curl/legacy) is unchanged. See `docs/reliable-input-delivery.md`.
+
+  **Gesture beta: drag agent windows.** With the camera hand-tracking overlay, you can now pinch and move the floating subagent and ultracode run/transcript windows. They keep their glowing connector line to the session tab while moving and can travel across a multi-monitor seam.
+
+## 1.1.15
+
+### Patch Changes
+
+- Security: harden all frontend inline `onclick`/`ondblclick` handlers against a stored-XSS double-context bug.
+
+  Many inline handlers interpolated values as `'${escapeHtml(value)}'` — a JavaScript string literal sitting inside an HTML attribute. The browser HTML-decodes the attribute value _before_ parsing the handler source, so `escapeHtml`'s `&#39;` reverts to a literal `'` and a quote-bearing id/name/path/URL breaks out of the JS string into executable code. `escapeHtml` alone is insufficient for this JS-string-within-HTML-attribute context.
+
+  All affected handlers now use `escapeHtml(JSON.stringify(value))`: `JSON.stringify` JS-encodes and quote-wraps the value, then `escapeHtml` handles the HTML-attribute layer, so the value round-trips as a single inert string argument.
+  - ultracode run/agent cards and minimized-tab badges (`ultracode-panel.js`, `ultracode-windows.js`) — PR #132.
+  - Session tabs (click/rename/gear/detach/close), notifications, subagent windows + dropdowns, the agents/tools/log-viewer/image-popup panels, mux-session monitor rows, and case-management buttons (`app.js`, `notification-manager.js`, `subagent-windows.js`, `panels-ui.js`, `session-ui.js`).
+  - Two non-`escapeHtml` variants of the same class: a pre-escaped mux-session id in `panels-ui.js` (`selectSession`/`killMuxSession`) and a fully raw, unescaped `phase.id` in `orchestrator-panel.js` (`orchestratorSkipPhase`/`orchestratorRetryPhase`).
+
+  The most realistic exploitation vector was file paths in the project-insights log-viewer link, since filenames can legally contain a single quote. Purely numeric interpolations and developer-literal handler strings were left unchanged.
+
+## 1.1.14
+
+### Patch Changes
+
+- Ultracode (Workflow-tool) floating windows — agent transcripts in-page, and minimize-to-tab.
+  - **Agent transcripts open in-page, connected, instead of a detached browser popup.** Clicking an agent card (in a run window or the dock panel) now opens the agent's live transcript as its own draggable floating window, tied by a connector line to its parent run window (falling back to the run's session tab if that window has since closed) — the same line idiom the run windows use. Re-clicking a card focuses the existing window; closing it removes the window and its line. (Previously this spawned a separate `window.open` browser popup.)
+  - **The window "−" button now minimizes into the originating session tab**, mirroring the subagent-window idiom. The window genie-animates into its tab and is tracked there; the tab shows an `ULTRA` badge whose hover/click dropdown lists each minimized item (🧬 run windows, 📄 agent transcripts). Click an item to restore its floating window, or dismiss it with ×. A run minimized while still active keeps tracking in the background and its badge auto-clears shortly after the run finishes. Both run windows and agent-transcript windows minimize into the same merged badge.
+  - Removed the old collapse-to-header behavior that the "−" button previously triggered (now superseded by minimize-to-tab).
+
+## 1.1.13
+
+### Patch Changes
+
+- Keep the `/compact` button in the extended (full) mobile keyboard accessory bar; only the simple bar drops it. (1.1.12 had removed it from both.)
+
 ## 1.1.12
 
 ### Patch Changes

CLAUDE.md

@@ -56,7 +56,7 @@ When user says "COM":
 
 CI runs `npm run check:lockfile` on every push/PR, so lockfile drift fails the build even if the `version-packages` script is bypassed.
 
-**Version**: 1.1.12 (must match `package.json`)
+**Version**: 1.1.17 (must match `package.json`)
 
 ## Project Overview
 
@@ -218,11 +218,11 @@ Frontend JS modules have `@fileoverview` with `@dependency`/`@loadorder` tags. L
 
 ### SSE Event Registry
 
-~120 event types in `src/web/sse-events.ts` (backend) and `SSE_EVENTS` in `constants.js` (frontend). Both must be kept in sync.
+~127 event types in `src/web/sse-events.ts` (backend) and `SSE_EVENTS` in `constants.js` (frontend). Both must be kept in sync.
 
 ### API Routes
 
-~146 handlers across 16 route files in `src/web/routes/`: system (41, incl. self-update `check`/`status`/`POST /api/system/update`, `POST /api/system/span-displays` → spawns `scripts/span-codeman.sh`, and `GET /api/codex/status`), sessions (29), orchestrator (10), cases (9), ralph (9), plan (8), files (14, incl. attachment register + list/history + `:attachmentId/raw`/`preview`/`thumbnail` + workspace `file-preview`/`file-thumbnail`), respawn (7), mux (5), push (4), scheduled (4), teams (2), hooks (1), clipboard (1), status-telemetry (1, `POST /api/status-telemetry` ← statusLine exporter), ws (1 WebSocket). Each file has `@fileoverview` with endpoint details.
+~147 handlers across 16 route files in `src/web/routes/`: system (41, incl. self-update `check`/`status`/`POST /api/system/update`, `POST /api/system/span-displays` → spawns `scripts/span-codeman.sh`, and `GET /api/codex/status`), sessions (29), orchestrator (10), cases (9), ralph (9), plan (8), files (14, incl. attachment register + list/history + `:attachmentId/raw`/`preview`/`thumbnail` + workspace `file-preview`/`file-thumbnail`), respawn (7), mux (5), push (4), scheduled (4), teams (2), hooks (1), clipboard (1), status-telemetry (1, `POST /api/status-telemetry` ← statusLine exporter), ws (1 WebSocket). Each file has `@fileoverview` with endpoint details.
 
 **HTTP contract** (stable since 0.9.x, see `docs/versioning-policy.md`; full envelope/status/error-code/SSE spec in `docs/api-reference.md`): responses use the `ApiResponse<T>` envelope — `{ success: true, data? }` or `{ success: false, error, errorCode }` (`src/types/api.ts`). `/api/v1/*` is a versioned alias of `/api/*` (URL rewrite in `server.ts`).
 

docs/reliable-input-delivery.md

@@ -0,0 +1,72 @@
+# Reliable input delivery (exactly-once, durable)
+
+## The bug this fixes
+
+With local echo on, pressing Enter cleared the overlay and then sent the prompt
+over the WebSocket **fire-and-forget** (`ws.send({t:'i',d})`). On a flaky link
+(e.g. a moving train) the socket is frequently *half-open*: `readyState === OPEN`
+so `ws.send()` does **not** throw, but the underlying TCP is dead, so the frame is
+silently discarded. Nothing was enqueued (the send "succeeded"), the on-screen
+prompt was already wiped, and `navigator.onLine` stays `true` — so a long typed
+prompt vanished with no trace and no resend.
+
+## The guarantee
+
+Every byte of user input is **recorded durably before delivery** and **only
+dropped once the server ACKs it** — so a half-open socket, a reconnect, or a page
+reload can never lose input. Redelivery is **exactly-once**: the server applies
+each `(clientId, seq)` at most once, so a resend can't type the prompt twice.
+
+## How it works
+
+### Client (`app.js`)
+
+- A stable **`clientId`** (`localStorage['codeman:clientId']`) identifies this
+  browser to the server's dedup across reconnects and reloads.
+- Each input frame gets a **monotonic per-session `seq`**. Frame records
+  (`{seq,data,useMux,ts,tries,sentAt}`) live in `_pendingDeliveries`
+  (`Map<sessionId, record[]>`), persisted (debounced, + flushed on `pagehide`/
+  `visibilitychange`) to `localStorage['codeman:pendingInput']`. The seq counters
+  persist too, so seqs stay monotonic across reloads (never reset — a reset would
+  let the server treat fresh input as an already-applied duplicate).
+- **Delivery** (`_drainSession`):
+  - **WS path** — when the socket is `OPEN` for the session, send each not-yet-sent
+    record (`sentAt === 0`) in seq order over the single ordered stream. Records
+    stay pending until the server's `{t:'ia',seq}` ACK removes them.
+  - **POST path** — when no WS, POST records in order, awaiting each (the HTTP 2xx
+    *is* the ACK). A 404/410 (session gone) drops the record rather than retry
+    forever.
+- **Half-open recovery** (`_redeliverSweep`, every 2s): if the active WS session's
+  oldest record is unacked past `_reliableAckTimeoutMs` (4s), the socket is assumed
+  dead — `ws.close()` forces a fast reconnect; `onopen` (`_onWsReady`) resets
+  `sentAt = 0` and re-sends everything pending. Also re-drains background sessions
+  over POST, and fires on SSE-reconnect / `online`.
+- The connection indicator shows pending count/bytes (`_pendingBytes`).
+
+### Server
+
+- **`Session.shouldApplyInput(clientId, seq)`** — returns `true` exactly once per
+  `(clientId, seq)`: the first time a seq strictly greater than that client's
+  last-applied is seen. A replayed/lower seq returns `false`. Bounded MRU map
+  (`MAX_INPUT_DEDUP_CLIENTS = 256`).
+- **WS route** (`ws-routes.ts`) — parses optional `cid`/`seq` on `{t:'i'}`; applies
+  via `shouldApplyInput` (skips a duplicate, still ACKs with `{t:'ia',seq}` so the
+  client drops it). Untagged frames apply unconditionally (no behavior change).
+- **POST route** (`/api/sessions/:id/input`) — optional `seq`/`clientId` in
+  `SessionInputWithLimitSchema`; a deduped duplicate returns 200 without writing
+  (the 200 is the client's ACK). `curl`/legacy callers omit the fields and always
+  apply.
+
+## Known limitation
+
+Dedup state is in-memory on the server. A **server restart** between a write and
+the client's redelivery of that same seq could re-apply it (a rare duplicate).
+This is a deliberate trade-off: favor *never losing input* over a rare duplicate
+across the narrow restart window.
+
+## Tests
+
+- `test/reliable-input-dedup.test.ts` — `Session.shouldApplyInput` exactly-once
+  semantics (monotonic, per-client, gap-tolerant, eviction-safe).
+- `test/routes/session-routes.test.ts` — POST `/input` applies a tagged
+  `(clientId, seq)` once on redelivery; untagged input always applies.

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "aicodeman",
-  "version": "1.1.12",
+  "version": "1.1.17",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "aicodeman",
-      "version": "1.1.12",
+      "version": "1.1.17",
       "hasInstallScript": true,
       "license": "MIT",
       "workspaces": [

package.json

@@ -1,6 +1,6 @@
 {
   "name": "aicodeman",
-  "version": "1.1.12",
+  "version": "1.1.17",
   "description": "Mission control for AI coding agents - run 20 autonomous agents with real-time monitoring and session persistence",
   "type": "module",
   "main": "dist/index.js",

packages/gesture-control/src/codeman/entry.ts

@@ -17,6 +17,13 @@
 //   • Panel "re-grab" — pinch an existing floating panel and move it anywhere;
 //     release over the tab strip to re-dock it (panel goes away, the tab stays).
 //     This is the capability the old OS-window detach lost.
+//   • Agent-window "grab-to-move" — pinch any floating *subagent* or *ultracode*
+//     run/transcript window (the dashboard's own `.subagent-window` /
+//     `.ultracode-window` floats) and move it anywhere. These windows stay owned
+//     by app.js — we only nudge their `style.left/top` and ask app.js to redraw
+//     the glowing connector line back to their session tab (its redraw reads live
+//     rects, so the line tracks without us touching app.js internals). This is the
+//     multi-monitor verb that lets these windows cross the physical monitor seam.
 //   • Button "tap" — pinch over a toolbar button (Run / Run Shell) and release
 //     in place → fires the button's real click handler. Drift too far first and
 //     it's treated as a stray move, not a tap.
@@ -36,12 +43,29 @@ import type { HandState } from '../gesture/types.ts';
 declare global {
   interface Window {
     __codemanGesture?: GestureBridge;
+    /** The Codeman dashboard singleton (app.js, `window.app`). The gesture layer
+     *  reaches into it to redraw the floating-window connector lines and bump a
+     *  grabbed window's z-order while moving the subagent / ultracode windows.
+     *  Loosely typed — only the few members we touch. */
+    app?: {
+      updateConnectionLines?: () => void;
+      saveSubagentWindowStates?: () => void;
+      subagentWindowZIndex?: number;
+      ultracodeWindowZIndex?: number;
+    };
   }
 }
 
 const TAB_SELECTOR = '.session-tab';
 /** An in-page floating session panel this layer spawned — re-grabbable to move. */
 const PANEL_SELECTOR = '.cg-float';
+/** The dashboard's own floating agent windows (subagent runs + ultracode run and
+ *  transcript windows). All three carry one of these classes, position via
+ *  `style.left/top`, and redraw their connector line from
+ *  `window.app.updateConnectionLines()` — so the hand can pick one up and move it
+ *  without app.js knowing. (`.ultracode-agent-window` also carries
+ *  `.ultracode-window`, so this matches it too.) */
+const WINDOW_SELECTOR = '.subagent-window, .ultracode-window';
 /** The session-tab strip; dropping a moved panel over it re-docks the session. */
 const DOCK_SELECTOR = '.session-tabs';
 /** Toolbar buttons a pinch can "tap": Run (#runBtn → app.run()) and Run Shell
@@ -93,6 +117,17 @@ type Grab =
       dy: number;
       /** Cursor currently over the tab strip → releasing re-docks. */
       overDock: boolean;
+    }
+  | {
+      /** A dashboard-owned floating agent window (subagent / ultracode) being
+       *  moved. We never remove or re-parent it — just reposition + redraw its
+       *  connector. The element ref can go stale mid-grab (SSE reconnect tears
+       *  ultracode windows down), so every move guards on `el.isConnected`. */
+      kind: 'window';
+      el: HTMLElement;
+      /** Cursor→window-top-left offset at grab, so it doesn't snap. */
+      dx: number;
+      dy: number;
     };
 
 /** Live state for one hand pinching a toolbar button (Run / Run Shell). */
@@ -122,6 +157,8 @@ class GestureBridge {
   private taps = new Map<string, Tap>();
   /** Live floating panels, keyed by session id (idempotent per id). */
   private floats = new Map<string, FloatingPanel>();
+  /** rAF coalescing for connector-line redraws while dragging an agent window. */
+  private connectorRedrawScheduled = false;
 
   constructor() {
     injectStyles();
@@ -187,7 +224,7 @@ class GestureBridge {
       await this.gc.start();
       this.running = true;
       this.button.classList.add('on');
-      this.status.textContent = 'on — pinch a tab or button';
+      this.status.textContent = 'on — pinch a tab, window, or button';
     } catch (err) {
       // Surface the *real* cause: MediaPipe/Emscripten can throw a non-Error
       // (number/string), so `(err as Error).message` was logging "undefined".
@@ -242,6 +279,22 @@ class GestureBridge {
       }
     }
 
+    // A dashboard-owned floating agent window (subagent / ultracode run or
+    // transcript) → pick it up and move it. Priority below cg-float panels
+    // (which sit far above), above tabs/buttons. We grab anywhere on the window
+    // (not just its titlebar) since the hand is choosing the whole window.
+    const win = this.hitClosest(x, y, WINDOW_SELECTOR);
+    if (win) {
+      const rect = win.getBoundingClientRect();
+      // Match app.js's own drag: drop any bottom-anchor so left/top take effect.
+      win.style.bottom = 'auto';
+      win.classList.add('cg-win-grabbed');
+      this.bringWindowToFront(win);
+      this.grabs.set(hand, { kind: 'window', el: win, dx: x - rect.left, dy: y - rect.top });
+      this.status.textContent = 'moving window';
+      return;
+    }
+
     // A session tab → grab-and-pull-out into a floating panel (ghost follows).
     const tab = this.hitClosest(x, y, TAB_SELECTOR);
     const id = tab?.dataset.id;
@@ -292,12 +345,16 @@ class GestureBridge {
       }
       return;
     }
+    if (grab?.kind === 'window') {
+      this.moveWindow(grab.el, x - grab.dx, y - grab.dy);
+      return;
+    }
     // A button pinch that drifts too far is a stray move, not a tap — cancel it.
     const tap = this.taps.get(hand);
     if (tap && Math.hypot(x - tap.ox, y - tap.oy) > TAP_CANCEL_PX) {
       tap.el.classList.remove('cg-tap-armed');
       this.taps.delete(hand);
-      this.status.textContent = 'on — pinch a tab or button';
+      this.status.textContent = 'on — pinch a tab, window, or button';
     }
   }
 
@@ -319,6 +376,23 @@ class GestureBridge {
       else this.flash('placed');
       return;
     }
+    if (grab?.kind === 'window') {
+      this.grabs.delete(hand);
+      grab.el.classList.remove('cg-win-grabbed');
+      // Clear the coalescer so the final placement always redraws, even if a
+      // mid-drag rAF was throttled (tab briefly backgrounded) and left it latched.
+      this.connectorRedrawScheduled = false;
+      this.redrawWindowConnectors();
+      // Persist subagent-window positions like app.js's own drag end does
+      // (a no-op for ultracode windows, which aren't position-persisted).
+      try {
+        window.app?.saveSubagentWindowStates?.();
+      } catch {
+        /* best-effort */
+      }
+      this.flash('placed window');
+      return;
+    }
     // Release over the same button → fire its real click handler.
     const tap = this.taps.get(hand);
     if (tap) {
@@ -373,6 +447,59 @@ class GestureBridge {
     float.el.style.top = `${t}px`;
   }
 
+  /** Move a dashboard-owned agent window by its top-left, clamped on-screen, then
+   *  redraw its connector line. The window self-positions via `style.left/top` and
+   *  app.js's connector redraw reads live rects, so this tracks without touching
+   *  app.js internals. Guards on `isConnected`: ultracode windows can be torn down
+   *  (SSE reconnect / auto-close) while still held. Clamps to `innerWidth/Height`,
+   *  which equals the *spanned* viewport in a multi-monitor window — so the window
+   *  can still travel across the physical monitor seam, just not off-screen. */
+  private moveWindow(el: HTMLElement, left: number, top: number): void {
+    if (!el.isConnected) return;
+    const w = el.offsetWidth || 380;
+    const h = el.offsetHeight || 320;
+    const l = Math.min(Math.max(4, left), Math.max(4, window.innerWidth - w - 4));
+    const t = Math.min(Math.max(4, top), Math.max(4, window.innerHeight - h - 4));
+    el.style.left = `${l}px`;
+    el.style.top = `${t}px`;
+    this.redrawWindowConnectors();
+  }
+
+  /** Ask app.js to redraw all connector lines (subagent + ultracode), coalesced to
+   *  one per frame so per-frame drags don't thrash. `updateConnectionLines()` is
+   *  itself debounced in app.js, but we rAF-gate too in case an older dashboard
+   *  build isn't, and to no-op cleanly when app.js isn't present (standalone). */
+  private redrawWindowConnectors(): void {
+    if (this.connectorRedrawScheduled) return;
+    this.connectorRedrawScheduled = true;
+    requestAnimationFrame(() => {
+      this.connectorRedrawScheduled = false;
+      try {
+        window.app?.updateConnectionLines?.();
+      } catch {
+        /* app.js may not expose it (standalone playground) */
+      }
+    });
+  }
+
+  /** Pop a grabbed window above its siblings using app.js's own z-counter, so a
+   *  picked-up window comes to the front like a real focus. Cosmetic + best-effort. */
+  private bringWindowToFront(el: HTMLElement): void {
+    const app = window.app;
+    if (!app) return;
+    try {
+      if (el.classList.contains('ultracode-window')) {
+        app.ultracodeWindowZIndex = (app.ultracodeWindowZIndex ?? 1000) + 1;
+        el.style.zIndex = String(app.ultracodeWindowZIndex);
+      } else {
+        app.subagentWindowZIndex = (app.subagentWindowZIndex ?? 1000) + 1;
+        el.style.zIndex = String(app.subagentWindowZIndex);
+      }
+    } catch {
+      /* cosmetic only */
+    }
+  }
+
   private positionGhost(ghost: HTMLElement, x: number, y: number): void {
     ghost.style.left = `${x}px`;
     ghost.style.top = `${y}px`;
@@ -385,17 +512,19 @@ class GestureBridge {
       if (grab.kind === 'tab') {
         grab.ghost.remove();
         grab.tab.classList.remove('cg-grabbed');
-      } else {
+      } else if (grab.kind === 'panel') {
         grab.panel.el.style.pointerEvents = '';
         grab.panel.el.classList.remove('cg-float-grabbed', 'cg-redock');
+      } else {
+        grab.el.classList.remove('cg-win-grabbed');
       }
     }
     this.grabs.clear();
     for (const tap of this.taps.values()) tap.el.classList.remove('cg-tap-armed');
     this.taps.clear();
     document
-      .querySelectorAll(`${TAB_SELECTOR}.cg-grabbed, .cg-tap-armed`)
-      .forEach((t) => t.classList.remove('cg-grabbed', 'cg-tap-armed'));
+      .querySelectorAll(`${TAB_SELECTOR}.cg-grabbed, .cg-tap-armed, .cg-win-grabbed`)
+      .forEach((t) => t.classList.remove('cg-grabbed', 'cg-tap-armed', 'cg-win-grabbed'));
   }
 
   private onStatus(fps: number, hands: HandState[]): void {
@@ -491,6 +620,10 @@ function injectStyles(): void {
   .cg-status { color: #9aa0a6; max-width: 220px; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
   .session-tab.cg-grabbed { opacity: .35; outline: 2px dashed #4ade80; outline-offset: -2px; }
   .cg-tap-armed { outline: 2px solid #4ade80 !important; outline-offset: 2px; box-shadow: 0 0 0 4px rgba(74,222,128,.25) !important; }
+  .subagent-window.cg-win-grabbed, .ultracode-window.cg-win-grabbed {
+    outline: 2px solid #4ade80 !important; outline-offset: -2px;
+    box-shadow: 0 12px 48px rgba(74,222,128,.5) !important;
+  }
   .cg-float {
     position: fixed; left: 0; top: 0; width: ${FLOAT_W}px; height: ${FLOAT_H}px;
     z-index: ${Z}; display: flex; flex-direction: column; overflow: hidden;

src/config/buffer-limits.ts

@@ -96,3 +96,18 @@ export const TRIM_RESPAWN_BUFFER_TO = 512 * 1024; // 512KB
  * which is enough to extract metadata from the first few JSONL lines.
  */
 export const FILE_PEEK_BYTES = 8 * 1024 - 1; // 8KB (inclusive end offset)
+
+// ============================================================================
+// Paste-Image Upload Limits
+// ============================================================================
+
+/**
+ * Maximum size (bytes) of a single image uploaded via POST
+ * /api/sessions/:id/paste-image. The mobile picker / drag-drop / paste paths
+ * send one file per request (the client uploads up to MAX_PASTE_IMAGES of them
+ * per batch), so this caps each individual file, not the batch. Generous enough
+ * for full-resolution phone photos and large screenshots; the client downscales
+ * very large images before upload, so legitimate uploads land well under this.
+ * Override: CODEMAN_MAX_PASTE_IMAGE_BYTES (bytes)
+ */
+export const MAX_PASTE_IMAGE_BYTES = parseInt(process.env.CODEMAN_MAX_PASTE_IMAGE_BYTES || '') || 50 * 1024 * 1024; // 50MB

src/session.ts

@@ -2213,6 +2213,42 @@ export class Session extends EventEmitter {
     }
   }
 
+  /**
+   * Per-client highest-applied input sequence, for exactly-once input delivery.
+   * Keyed by the web client's stable `clientId`. Bounded so many devices over a
+   * long-lived session can't grow it without limit (insertion order = MRU, so
+   * eviction drops the least-recently-active client).
+   */
+  private _appliedInputSeq = new Map<string, number>();
+  private static readonly MAX_INPUT_DEDUP_CLIENTS = 256;
+
+  /**
+   * Decide whether an input frame should be applied to the PTY or skipped as a
+   * duplicate redelivery. Returns true exactly once per (clientId, seq): the
+   * first time a seq strictly greater than the client's last-applied is seen.
+   * A redelivery of an already-applied seq (the client never got our ACK and
+   * resent) returns false. Callers should ACK regardless — a duplicate is, from
+   * the client's view, "delivered" — and only `write()` the PTY when this is
+   * true. Relies on the client delivering one client's frames in seq order over
+   * a single ordered stream, so `seq <= last` ⇒ already applied.
+   *
+   * Without this, the client's at-least-once redelivery (needed because a
+   * half-open socket silently drops frames with no error) would type a prompt
+   * twice whenever an ACK is lost after the write landed.
+   */
+  shouldApplyInput(clientId: string, seq: number): boolean {
+    const last = this._appliedInputSeq.get(clientId);
+    if (last !== undefined && seq <= last) return false;
+    // Re-insert to move this client to the MRU end for fair eviction.
+    if (last !== undefined) this._appliedInputSeq.delete(clientId);
+    this._appliedInputSeq.set(clientId, seq);
+    if (this._appliedInputSeq.size > Session.MAX_INPUT_DEDUP_CLIENTS) {
+      const oldest = this._appliedInputSeq.keys().next().value;
+      if (oldest !== undefined) this._appliedInputSeq.delete(oldest);
+    }
+    return true;
+  }
+
   /**
    * Sends input via the terminal multiplexer's direct input mechanism.
    *

src/web/public/app.js

@@ -468,13 +468,28 @@ class CodemanApp {
     this.maxReconnectAttempts = 10;
     this.isOnline = navigator.onLine;
 
-    // Offline input queue
-    this._inputQueue = new Map(); // Map<sessionId, string>
-    this._inputQueueMaxBytes = 64 * 1024; // 64KB cap per session
+    // Reliable, durable input delivery (replaces the old best-effort queue).
+    // Every input byte is recorded with a stable clientId + a monotonic
+    // per-session seq, persisted to localStorage, and only dropped once the
+    // server ACKs that exact seq — so a half-open socket silently dropping a
+    // frame, a reconnect, or a page reload can never lose a typed prompt.
+    // Exactly-once: the server applies each (clientId, seq) at most once.
     this._connectionStatus = 'connected';
-
-    // Sequential input send chain — ensures keystroke ordering across async fetches
-    this._inputSendChain = Promise.resolve();
+    this._clientId = '';
+    this._seqCounters = new Map(); // sessionId -> last issued seq
+    this._pendingDeliveries = new Map(); // sessionId -> [{seq,data,useMux,ts,tries,sentAt}]
+    this._postDraining = new Set(); // sessionIds with an in-flight POST drainer
+    this._persistReliableTimer = null;
+    this._reliableAckTimeoutMs = 4000; // unacked WS frame older than this ⇒ socket likely dead
+    this._reliableMaxBytes = 256 * 1024; // cap on the persisted backlog
+    this._loadReliableState();
+    this._reliableSweepTimer = setInterval(() => this._redeliverSweep(), 2000);
+    // Flush the durable queue synchronously when the page is hidden/closed —
+    // debounced persistence may have a pending write we mustn't lose on reload.
+    window.addEventListener('pagehide', () => this._persistReliableNow());
+    document.addEventListener('visibilitychange', () => {
+      if (document.visibilityState === 'hidden') this._persistReliableNow();
+    });
 
     // Local echo overlay — DOM overlay positioned at the visible ❯ prompt
     // (not at buffer.cursorY, which reflects Ink's internal cursor position)
@@ -1931,8 +1946,10 @@ class CodemanApp {
   setConnectionStatus(status) {
     this._connectionStatus = status;
     this._updateConnectionIndicator();
-    if (status === 'connected' && this._inputQueue.size > 0) {
-      this._drainInputQueues();
+    if (status === 'connected') {
+      // Reconnected (SSE) — push any durably-queued input out immediately
+      // instead of waiting for the next 2s sweep.
+      this._redeliverSweep();
     }
   }
 
@@ -1965,6 +1982,9 @@ class CodemanApp {
         // went over HTTP, which never claims (see ws-routes sizingToken).
         this.sendResize(sessionId)?.catch?.(() => {});
         this._startMobileResizeRetry(sessionId);
+        // Flush any durably-queued input over the fresh socket (covers frames a
+        // prior half-open socket silently dropped, and input typed while offline).
+        this._onWsReady(sessionId);
       }
     };
 
@@ -1979,6 +1999,10 @@ class CodemanApp {
           this._onSessionClearTerminal({ id: sessionId });
         } else if (msg.t === 'r') {
           this._onSessionNeedsRefresh({ id: sessionId });
+        } else if (msg.t === 'ia') {
+          // Input ACK — the server applied (or deduped) this seq; drop it from
+          // the durable queue so it can never be re-delivered/lost.
+          this._onWsInputAck(msg.seq);
         }
       } catch {
         // Ignore malformed messages
@@ -2062,79 +2086,270 @@ class CodemanApp {
   }
 
   /**
-   * Send input to server without blocking the keystroke flush cycle.
-   * Uses a sequential promise chain to preserve character ordering
-   * across concurrent async fetches.
+   * Public input entry point — name/signature kept for all call sites.
+   * Records the input durably, then delivers it reliably (exactly-once). Never
+   * blocks the keystroke flush; never silently drops on a half-open socket.
+   * @param {string} sessionId
+   * @param {string} input
+   * @param {{useMux?: boolean}} [opts] - useMux only affects the POST fallback.
    */
-  _sendInputAsync(sessionId, input) {
-    // Queue immediately if offline
-    if (!this.isOnline || this._connectionStatus === 'disconnected') {
-      this._enqueueInput(sessionId, input);
+  _sendInputAsync(sessionId, input, opts) {
+    if (!sessionId || !input) return;
+    this._reliableSend(sessionId, input, opts?.useMux === true);
+  }
+
+  /** Record one input frame and kick delivery. The record lives until ACKed. */
+  _reliableSend(sessionId, data, useMux) {
+    const seq = this._nextSeq(sessionId);
+    const rec = { seq, data, useMux: !!useMux, ts: Date.now(), tries: 0, sentAt: 0 };
+    let list = this._pendingDeliveries.get(sessionId);
+    if (!list) {
+      list = [];
+      this._pendingDeliveries.set(sessionId, list);
+    }
+    list.push(rec);
+    this._persistReliableState();
+    this._updateConnectionIndicator();
+    this._drainSession(sessionId);
+  }
+
+  _nextSeq(sessionId) {
+    const next = (this._seqCounters.get(sessionId) || 0) + 1;
+    this._seqCounters.set(sessionId, next);
+    return next;
+  }
+
+  /** Deliver all unacked records for a session, in seq order. */
+  _drainSession(sessionId) {
+    const list = this._pendingDeliveries.get(sessionId);
+    if (!list || list.length === 0) return;
+
+    // Fast path: WebSocket open for this session — fire each not-yet-sent record
+    // over the single ordered stream. They stay pending until the server ACKs
+    // them ({t:'ia'}); a frame swallowed by a half-open socket is re-sent after
+    // the sweep force-reconnects (which resets sentAt=0 in _onWsReady).
+    if (this._ws && this._ws.readyState === WebSocket.OPEN && this._wsSessionId === sessionId) {
+      for (const rec of list) {
+        if (rec.sentAt !== 0) continue;
+        try {
+          this._ws.send(JSON.stringify({ t: 'i', d: rec.data, seq: rec.seq, cid: this._clientId }));
+          rec.sentAt = Date.now();
+          rec.tries++;
+        } catch {
+          break; // socket died mid-send — reconnect/POST drainer retries
+        }
+      }
       return;
     }
 
-    // Fast path: WebSocket — fire-and-forget, inherently ordered (single TCP stream).
-    if (this._wsReady && this._wsSessionId === sessionId) {
+    // Slow path: no WS — POST records in order, awaiting each (the HTTP 2xx is
+    // the ACK). Serialized per session so seq order survives async fetches.
+    if (this._postDraining.has(sessionId)) return;
+    this._postDraining.add(sessionId);
+    (async () => {
       try {
-        this._ws.send(JSON.stringify({ t: 'i', d: input }));
-        this.clearPendingHooks(sessionId);
-        return;
-      } catch {
-        // WS send failed — fall through to HTTP POST
-      }
-    }
-
-    // Slow path: HTTP POST — chain on dispatch only, don't wait for response.
-    // The server handles writeViaMux as fire-and-forget anyway.
-    this._inputSendChain = this._inputSendChain.then(() => {
-      const fetchPromise = fetch(`/api/sessions/${sessionId}/input`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({ input }),
-        keepalive: input.length < 65536,
-      });
-
-      // Handle response asynchronously — don't block next keystroke on response
-      fetchPromise.then(resp => {
-        if (!resp.ok) {
-          this._enqueueInput(sessionId, input);
-        } else {
-          this.clearPendingHooks(sessionId);
+        for (;;) {
+          const cur = this._pendingDeliveries.get(sessionId);
+          if (!cur || cur.length === 0) break;
+          // If the WebSocket came back mid-drain, yield to it (the acked stream)
+          // so we don't redundantly re-POST what onopen is already re-sending.
+          if (this._ws && this._ws.readyState === WebSocket.OPEN && this._wsSessionId === sessionId) {
+            break;
+          }
+          const rec = cur[0];
+          rec.tries++;
+          rec.sentAt = Date.now();
+          let resp = null;
+          try {
+            const body = { input: rec.data, seq: rec.seq, clientId: this._clientId };
+            if (rec.useMux) body.useMux = true;
+            resp = await fetch(`/api/sessions/${sessionId}/input`, {
+              method: 'POST',
+              headers: { 'Content-Type': 'application/json' },
+              body: JSON.stringify(body),
+              keepalive: rec.data.length < 65536,
+            });
+          } catch {
+            resp = null;
+          }
+          if (resp && resp.ok) {
+            this._ackDelivery(sessionId, rec.seq);
+          } else if (resp && (resp.status === 404 || resp.status === 410)) {
+            // Session no longer exists — the input can never land. Drop it
+            // rather than retry forever (not a "lost" prompt: the target is gone).
+            this._ackDelivery(sessionId, rec.seq);
+          } else {
+            break; // offline / 5xx — leave queued; sweep + reconnect retry later
+          }
         }
-      }).catch(() => {
-        this._enqueueInput(sessionId, input);
-      });
-
-      // Return immediately after fetch is dispatched (don't await response)
-    });
+      } finally {
+        this._postDraining.delete(sessionId);
+      }
+    })();
   }
 
-
-  _enqueueInput(sessionId, input) {
-    const existing = this._inputQueue.get(sessionId) || '';
-    let combined = existing + input;
-    // Enforce 64KB cap — keep most recent keystrokes
-    if (combined.length > this._inputQueueMaxBytes) {
-      combined = combined.slice(combined.length - this._inputQueueMaxBytes);
-    }
-    this._inputQueue.set(sessionId, combined);
-    this._updateConnectionIndicator();
-  }
-
-  async _drainInputQueues() {
-    if (this._inputQueue.size === 0) return;
-    // Snapshot and clear
-    const queued = new Map(this._inputQueue);
-    this._inputQueue.clear();
-    this._updateConnectionIndicator();
-
-    for (const [sessionId, input] of queued) {
-      const resp = await this._apiPost(`/api/sessions/${sessionId}/input`, { input });
-      if (!resp?.ok) {
-        this._enqueueInput(sessionId, input);
+  /** Drop an ACKed record (by exact seq) and persist. */
+  _ackDelivery(sessionId, seq) {
+    const list = this._pendingDeliveries.get(sessionId);
+    if (list) {
+      const idx = list.findIndex((r) => r.seq === seq);
+      if (idx !== -1) {
+        list.splice(idx, 1);
+        if (list.length === 0) this._pendingDeliveries.delete(sessionId);
+        // When nothing is left pending anywhere, flush durable state immediately
+        // (not debounced) so a reload in the next 250ms can't redeliver an
+        // already-delivered frame — otherwise localStorage briefly still shows it.
+        if (this._pendingDeliveries.size === 0) this._persistReliableNow();
+        else this._persistReliableState();
+        this._updateConnectionIndicator();
       }
     }
-    this._updateConnectionIndicator();
+    this.clearPendingHooks?.(sessionId);
+  }
+
+  /** Server input-ACK frame ({t:'ia',seq}) over the WebSocket. */
+  _onWsInputAck(seq) {
+    if (this._wsSessionId && Number.isInteger(seq)) this._ackDelivery(this._wsSessionId, seq);
+  }
+
+  /** Called from ws.onopen — flush everything pending over the fresh socket. */
+  _onWsReady(sessionId) {
+    const list = this._pendingDeliveries.get(sessionId);
+    if (list) for (const r of list) r.sentAt = 0; // fresh socket ⇒ re-send all
+    this._drainSession(sessionId);
+  }
+
+  /**
+   * Periodic retry. For the active WS session, an oldest frame unacked past the
+   * timeout means the socket is (half-)dead — close it to force a fast reconnect
+   * (onclose → reconnect → onopen → _onWsReady re-sends). Other sessions just
+   * (re)drain over POST.
+   */
+  _redeliverSweep() {
+    if (this._pendingDeliveries.size === 0) return;
+    for (const sessionId of [...this._pendingDeliveries.keys()]) {
+      const list = this._pendingDeliveries.get(sessionId);
+      if (!list || list.length === 0) continue;
+      const isActiveWs =
+        this._ws && this._ws.readyState === WebSocket.OPEN && this._wsSessionId === sessionId;
+      if (isActiveWs) {
+        const oldest = list[0];
+        if (oldest && oldest.sentAt && Date.now() - oldest.sentAt > this._reliableAckTimeoutMs) {
+          try {
+            this._ws.close(); // half-open: never recovers on its own — force reconnect
+          } catch {
+            /* ignore */
+          }
+          continue;
+        }
+      }
+      this._drainSession(sessionId);
+    }
+  }
+
+  /** Total bytes/count still awaiting ACK across all sessions (for the indicator). */
+  _pendingBytes() {
+    let bytes = 0;
+    let count = 0;
+    for (const list of this._pendingDeliveries.values()) {
+      for (const r of list) {
+        bytes += r.data.length;
+        count++;
+      }
+    }
+    return { bytes, count };
+  }
+
+  // ---- durable persistence (localStorage; quota- and disabled-storage-safe) --
+
+  _loadReliableState() {
+    // Stable client identity for server-side dedup across reconnects/reloads.
+    try {
+      this._clientId = localStorage.getItem('codeman:clientId') || '';
+    } catch {
+      this._clientId = '';
+    }
+    if (!this._clientId) {
+      this._clientId = 'c-' + Math.random().toString(36).slice(2) + '-' + Date.now().toString(36);
+      try {
+        localStorage.setItem('codeman:clientId', this._clientId);
+      } catch {
+        /* storage disabled — dedup degrades to per-load, still no loss */
+      }
+    }
+    try {
+      const raw = localStorage.getItem('codeman:pendingInput');
+      if (!raw) return;
+      const saved = JSON.parse(raw);
+      if (saved && saved.seqs) {
+        for (const [s, n] of Object.entries(saved.seqs)) {
+          if (Number.isFinite(n)) this._seqCounters.set(s, n);
+        }
+      }
+      if (saved && saved.pending) {
+        for (const [s, recs] of Object.entries(saved.pending)) {
+          if (Array.isArray(recs) && recs.length) {
+            // Reset sentAt so they re-deliver promptly on this fresh load.
+            this._pendingDeliveries.set(
+              s,
+              recs
+                .filter((r) => r && typeof r.data === 'string' && Number.isInteger(r.seq))
+                .map((r) => ({
+                  seq: r.seq,
+                  data: r.data,
+                  useMux: !!r.useMux,
+                  ts: r.ts || Date.now(),
+                  tries: 0,
+                  sentAt: 0,
+                }))
+            );
+          }
+        }
+      }
+    } catch {
+      /* corrupt/parse error — start clean rather than throw */
+    }
+  }
+
+  _persistReliableState() {
+    // Debounced — typing without local echo calls this per keystroke.
+    if (this._persistReliableTimer) return;
+    this._persistReliableTimer = setTimeout(() => {
+      this._persistReliableTimer = null;
+      this._persistReliableNow();
+    }, 250);
+  }
+
+  _persistReliableNow() {
+    if (this._persistReliableTimer) {
+      clearTimeout(this._persistReliableTimer);
+      this._persistReliableTimer = null;
+    }
+    try {
+      const seqs = {};
+      for (const [s, n] of this._seqCounters) seqs[s] = n;
+      const pending = {};
+      let bytes = 0;
+      for (const [s, list] of this._pendingDeliveries) {
+        if (!list.length) continue;
+        pending[s] = list.map((r) => ({
+          seq: r.seq,
+          data: r.data,
+          useMux: r.useMux,
+          ts: r.ts,
+          tries: r.tries,
+        }));
+        for (const r of list) bytes += r.data.length;
+      }
+      // Bound the persisted backlog. On extreme overflow keep the seq counters
+      // (so future input stays monotonic and dedup-safe) but skip the payloads —
+      // the in-memory queue still delivers; only cross-reload durability is lost.
+      const payload =
+        bytes > this._reliableMaxBytes ? { seqs } : { seqs, pending };
+      localStorage.setItem('codeman:pendingInput', JSON.stringify(payload));
+    } catch {
+      /* QuotaExceeded or disabled storage — in-memory delivery is unaffected */
+    }
   }
 
   _updateConnectionIndicator() {
@@ -2143,28 +2358,27 @@ class CodemanApp {
     const text = this.$('connectionText');
     if (!indicator || !dot || !text) return;
 
-    let totalBytes = 0;
-    for (const v of this._inputQueue.values()) totalBytes += v.length;
-
     const status = this._connectionStatus;
-    const hasQueue = totalBytes > 0;
 
-    // Connected with empty queue — hide
-    if ((status === 'connected' || status === 'connecting') && !hasQueue) {
+    // While the connection is healthy, never surface the input queue. With the
+    // reliable-delivery layer every keystroke is briefly "pending" until its ACK
+    // lands a few ms later — showing that flashed "Sending 1B…" on every single
+    // character. The indicator is only meaningful for an actual connection
+    // problem (reconnecting / offline), where the queued byte count reassures
+    // the user their typing is safely buffered and will be sent.
+    if (status === 'connected' || status === 'connecting') {
       indicator.style.display = 'none';
       return;
     }
 
+    const { bytes: totalBytes, count } = this._pendingBytes();
+    const hasQueue = count > 0;
     indicator.style.display = 'flex';
     dot.className = 'connection-dot';
 
-    const formatBytes = (b) => b < 1024 ? `${b}B` : `${(b / 1024).toFixed(1)}KB`;
+    const formatBytes = (b) => (b < 1024 ? `${b}B` : `${(b / 1024).toFixed(1)}KB`);
 
-    if (status === 'connected' && hasQueue) {
-      // Draining
-      dot.classList.add('draining');
-      text.textContent = `Sending ${formatBytes(totalBytes)}...`;
-    } else if (status === 'reconnecting') {
+    if (status === 'reconnecting') {
       dot.classList.add('reconnecting');
       text.textContent = hasQueue ? `Reconnecting (${formatBytes(totalBytes)} queued)` : 'Reconnecting...';
     } else {
@@ -2179,6 +2393,8 @@ class CodemanApp {
       this.isOnline = true;
       this.reconnectAttempts = 0;
       this.connectSSE();
+      // Network came back — drain durably-queued input right away.
+      this._redeliverSweep();
     });
     window.addEventListener('offline', () => {
       this.isOnline = false;
@@ -2773,12 +2989,16 @@ class CodemanApp {
       const minimizedCount = minimizedAgents?.size || 0;
       const subagentBadge = minimizedCount > 0 ? this.renderSubagentTabBadge(id, minimizedAgents) : '';
 
+      // Ultracode runs + agent transcripts minimized to this tab (ultracode-windows.js
+      // renders one merged ULTRA badge; returns '' when nothing is minimized).
+      const ultracodeBadge = this.renderUltracodeTabBadge ? this.renderUltracodeTabBadge(id) : '';
+
       // Show folder name if session has a custom name AND tall tabs setting is enabled
       const folderName = session.workingDir ? session.workingDir.split('/').pop() || '' : '';
       const tallTabsEnabled = this._tallTabsEnabled ?? false;
       const showFolder = tallTabsEnabled && session.name && folderName && folderName !== name;
 
-      parts.push(`<div class="session-tab ${isActive ? 'active' : ''}${alertClass}${loadState ? ' tab-loading' : ''}" data-id="${id}" data-color="${color}" ${loadState ? `data-load-phase="${escapeHtml(loadState.phase)}"` : ''} onclick="app.handleSessionTabClick(event, '${escapeHtml(id)}')" oncontextmenu="event.preventDefault(); app.startInlineRename('${escapeHtml(id)}')" tabindex="0" role="tab" aria-selected="${isActive ? 'true' : 'false'}" aria-busy="${loadState ? 'true' : 'false'}" aria-label="${escapeHtml(name)} session" ${session.workingDir ? `title="${escapeHtml(session.workingDir)}"` : ''}>
+      parts.push(`<div class="session-tab ${isActive ? 'active' : ''}${alertClass}${loadState ? ' tab-loading' : ''}" data-id="${id}" data-color="${color}" ${loadState ? `data-load-phase="${escapeHtml(loadState.phase)}"` : ''} onclick="app.handleSessionTabClick(event, ${escapeHtml(JSON.stringify(id))})" oncontextmenu="event.preventDefault(); app.startInlineRename(${escapeHtml(JSON.stringify(id))})" tabindex="0" role="tab" aria-selected="${isActive ? 'true' : 'false'}" aria-busy="${loadState ? 'true' : 'false'}" aria-label="${escapeHtml(name)} session" ${session.workingDir ? `title="${escapeHtml(session.workingDir)}"` : ''}>
           ${_tabIdx < 9 ? '<span class="tab-number">' + (_tabIdx + 1) + '</span>' : ''}
           ${loadState ? '<span class="tab-load-spinner" aria-hidden="true"></span>' : ''}
           <span class="tab-status ${status}" aria-hidden="true"></span>
@@ -2792,9 +3012,10 @@ class CodemanApp {
           </span>
           ${hasRunningTasks ? `<span class="tab-badge" onclick="event.stopPropagation(); app.toggleTaskPanel()" aria-label="${taskStats.running} running tasks">${taskStats.running}</span>` : ''}
           ${subagentBadge}
-          <span class="tab-gear" onclick="event.stopPropagation(); app.openSessionOptions('${escapeHtml(id)}')" title="Session options" aria-label="Session options" tabindex="0">&#x2699;</span>
-          <span class="tab-detach" onclick="event.stopPropagation(); app.detachSession('${escapeHtml(id)}')" title="Open in a new window" aria-label="Open session in a new window" tabindex="0">&#x29C9;</span>
-          <span class="tab-close" onclick="event.stopPropagation(); app.requestCloseSession('${escapeHtml(id)}')" title="Close session" aria-label="Close session" tabindex="0">&times;</span>
+          ${ultracodeBadge}
+          <span class="tab-gear" onclick="event.stopPropagation(); app.openSessionOptions(${escapeHtml(JSON.stringify(id))})" title="Session options" aria-label="Session options" tabindex="0">&#x2699;</span>
+          <span class="tab-detach" onclick="event.stopPropagation(); app.detachSession(${escapeHtml(JSON.stringify(id))})" title="Open in a new window" aria-label="Open session in a new window" tabindex="0">&#x29C9;</span>
+          <span class="tab-close" onclick="event.stopPropagation(); app.requestCloseSession(${escapeHtml(JSON.stringify(id))})" title="Close session" aria-label="Close session" tabindex="0">&times;</span>
         </div>`);
       _tabIdx++;
     }
@@ -3681,7 +3902,13 @@ class CodemanApp {
 
     this._flushedOffsets?.delete(sessionId);
     this._flushedTexts?.delete(sessionId);
-    this._inputQueue.delete(sessionId);
+    // Drop any durably-queued input for a session that's actually gone (deleted/
+    // exited). Not a lost prompt — the target no longer exists. Only reached on
+    // real session removal, never on a tab switch.
+    this._pendingDeliveries?.delete(sessionId);
+    this._seqCounters?.delete(sessionId);
+    this._postDraining?.delete(sessionId);
+    this._persistReliableState();
     this.ralphStates.delete(sessionId);
     this.ralphClosedSessions.delete(sessionId);
     this.projectInsights.delete(sessionId);

src/web/public/gesture/gesture-codeman.js

@@ -4449,6 +4449,7 @@ var GestureController = class {
 // packages/gesture-control/src/codeman/entry.ts
 var TAB_SELECTOR = ".session-tab";
 var PANEL_SELECTOR = ".cg-float";
+var WINDOW_SELECTOR = ".subagent-window, .ultracode-window";
 var DOCK_SELECTOR = ".session-tabs";
 var CLICK_SELECTOR = "#runBtn, .btn-shell";
 var Z2 = 2147483e3;
@@ -4476,6 +4477,8 @@ var GestureBridge = class {
     __publicField(this, "taps", /* @__PURE__ */ new Map());
     /** Live floating panels, keyed by session id (idempotent per id). */
     __publicField(this, "floats", /* @__PURE__ */ new Map());
+    /** rAF coalescing for connector-line redraws while dragging an agent window. */
+    __publicField(this, "connectorRedrawScheduled", false);
     injectStyles();
     this.surface = el("div", "cg-surface");
     this.canvas = el("canvas", "cg-canvas");
@@ -4530,7 +4533,7 @@ var GestureBridge = class {
       await this.gc.start();
       this.running = true;
       this.button.classList.add("on");
-      this.status.textContent = "on \u2014 pinch a tab or button";
+      this.status.textContent = "on \u2014 pinch a tab, window, or button";
     } catch (err) {
       const msg = describeError(err);
       this.status.textContent = `failed: ${msg}`;
@@ -4575,6 +4578,16 @@ var GestureBridge = class {
         return;
       }
     }
+    const win = this.hitClosest(x2, y2, WINDOW_SELECTOR);
+    if (win) {
+      const rect = win.getBoundingClientRect();
+      win.style.bottom = "auto";
+      win.classList.add("cg-win-grabbed");
+      this.bringWindowToFront(win);
+      this.grabs.set(hand, { kind: "window", el: win, dx: x2 - rect.left, dy: y2 - rect.top });
+      this.status.textContent = "moving window";
+      return;
+    }
     const tab = this.hitClosest(x2, y2, TAB_SELECTOR);
     const id = tab?.dataset.id;
     if (tab && id) {
@@ -4620,11 +4633,15 @@ var GestureBridge = class {
       }
       return;
     }
+    if (grab?.kind === "window") {
+      this.moveWindow(grab.el, x2 - grab.dx, y2 - grab.dy);
+      return;
+    }
     const tap = this.taps.get(hand);
     if (tap && Math.hypot(x2 - tap.ox, y2 - tap.oy) > TAP_CANCEL_PX) {
       tap.el.classList.remove("cg-tap-armed");
       this.taps.delete(hand);
-      this.status.textContent = "on \u2014 pinch a tab or button";
+      this.status.textContent = "on \u2014 pinch a tab, window, or button";
     }
   }
   onDrop(hand, x2, y2) {
@@ -4645,6 +4662,18 @@ var GestureBridge = class {
       else this.flash("placed");
       return;
     }
+    if (grab?.kind === "window") {
+      this.grabs.delete(hand);
+      grab.el.classList.remove("cg-win-grabbed");
+      this.connectorRedrawScheduled = false;
+      this.redrawWindowConnectors();
+      try {
+        window.app?.saveSubagentWindowStates?.();
+      } catch {
+      }
+      this.flash("placed window");
+      return;
+    }
     const tap = this.taps.get(hand);
     if (tap) {
       this.taps.delete(hand);
@@ -4694,6 +4723,54 @@ var GestureBridge = class {
     float.el.style.left = `${l}px`;
     float.el.style.top = `${t2}px`;
   }
+  /** Move a dashboard-owned agent window by its top-left, clamped on-screen, then
+   *  redraw its connector line. The window self-positions via `style.left/top` and
+   *  app.js's connector redraw reads live rects, so this tracks without touching
+   *  app.js internals. Guards on `isConnected`: ultracode windows can be torn down
+   *  (SSE reconnect / auto-close) while still held. Clamps to `innerWidth/Height`,
+   *  which equals the *spanned* viewport in a multi-monitor window — so the window
+   *  can still travel across the physical monitor seam, just not off-screen. */
+  moveWindow(el2, left, top) {
+    if (!el2.isConnected) return;
+    const w2 = el2.offsetWidth || 380;
+    const h2 = el2.offsetHeight || 320;
+    const l = Math.min(Math.max(4, left), Math.max(4, window.innerWidth - w2 - 4));
+    const t2 = Math.min(Math.max(4, top), Math.max(4, window.innerHeight - h2 - 4));
+    el2.style.left = `${l}px`;
+    el2.style.top = `${t2}px`;
+    this.redrawWindowConnectors();
+  }
+  /** Ask app.js to redraw all connector lines (subagent + ultracode), coalesced to
+   *  one per frame so per-frame drags don't thrash. `updateConnectionLines()` is
+   *  itself debounced in app.js, but we rAF-gate too in case an older dashboard
+   *  build isn't, and to no-op cleanly when app.js isn't present (standalone). */
+  redrawWindowConnectors() {
+    if (this.connectorRedrawScheduled) return;
+    this.connectorRedrawScheduled = true;
+    requestAnimationFrame(() => {
+      this.connectorRedrawScheduled = false;
+      try {
+        window.app?.updateConnectionLines?.();
+      } catch {
+      }
+    });
+  }
+  /** Pop a grabbed window above its siblings using app.js's own z-counter, so a
+   *  picked-up window comes to the front like a real focus. Cosmetic + best-effort. */
+  bringWindowToFront(el2) {
+    const app = window.app;
+    if (!app) return;
+    try {
+      if (el2.classList.contains("ultracode-window")) {
+        app.ultracodeWindowZIndex = (app.ultracodeWindowZIndex ?? 1e3) + 1;
+        el2.style.zIndex = String(app.ultracodeWindowZIndex);
+      } else {
+        app.subagentWindowZIndex = (app.subagentWindowZIndex ?? 1e3) + 1;
+        el2.style.zIndex = String(app.subagentWindowZIndex);
+      }
+    } catch {
+    }
+  }
   positionGhost(ghost, x2, y2) {
     ghost.style.left = `${x2}px`;
     ghost.style.top = `${y2}px`;
@@ -4703,15 +4780,17 @@ var GestureBridge = class {
       if (grab.kind === "tab") {
         grab.ghost.remove();
         grab.tab.classList.remove("cg-grabbed");
-      } else {
+      } else if (grab.kind === "panel") {
         grab.panel.el.style.pointerEvents = "";
         grab.panel.el.classList.remove("cg-float-grabbed", "cg-redock");
+      } else {
+        grab.el.classList.remove("cg-win-grabbed");
       }
     }
     this.grabs.clear();
     for (const tap of this.taps.values()) tap.el.classList.remove("cg-tap-armed");
     this.taps.clear();
-    document.querySelectorAll(`${TAB_SELECTOR}.cg-grabbed, .cg-tap-armed`).forEach((t2) => t2.classList.remove("cg-grabbed", "cg-tap-armed"));
+    document.querySelectorAll(`${TAB_SELECTOR}.cg-grabbed, .cg-tap-armed, .cg-win-grabbed`).forEach((t2) => t2.classList.remove("cg-grabbed", "cg-tap-armed", "cg-win-grabbed"));
   }
   onStatus(fps, hands) {
     const { width, height } = this.canvas;
@@ -4797,6 +4876,10 @@ function injectStyles() {
   .cg-status { color: #9aa0a6; max-width: 220px; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
   .session-tab.cg-grabbed { opacity: .35; outline: 2px dashed #4ade80; outline-offset: -2px; }
   .cg-tap-armed { outline: 2px solid #4ade80 !important; outline-offset: 2px; box-shadow: 0 0 0 4px rgba(74,222,128,.25) !important; }
+  .subagent-window.cg-win-grabbed, .ultracode-window.cg-win-grabbed {
+    outline: 2px solid #4ade80 !important; outline-offset: -2px;
+    box-shadow: 0 12px 48px rgba(74,222,128,.5) !important;
+  }
   .cg-float {
     position: fixed; left: 0; top: 0; width: ${FLOAT_W}px; height: ${FLOAT_H}px;
     z-index: ${Z2}; display: flex; flex-direction: column; overflow: hidden;

src/web/public/image-input.js

@@ -104,34 +104,78 @@ Object.assign(CodemanApp.prototype, {
     document.execCommand('paste');
   },
 
-  async _uploadAndInsertImages(files) {
+  // Max images accepted in one batch (paste / drop / mobile picker). Each is
+  // uploaded as its own request, so 20 stays under the server's 30 uploads/min
+  // rate limit while covering "select a bunch of photos at once".
+  _maxBatchImages: 20,
+  // How many uploads to run concurrently. Small enough that decoding several
+  // large images through <canvas> at once won't OOM a phone, large enough that
+  // 20 photos don't crawl through serially.
+  _uploadConcurrency: 3,
+
+  async _uploadAndInsertImages(fileList) {
     const sessionId = this.activeSessionId;
     if (!sessionId) return;
 
-    this.showToast('Uploading ' + files.length + ' image' + (files.length > 1 ? 's' : '') + '...', 'info');
+    let files = Array.from(fileList || []);
+    if (files.length === 0) return;
 
-    const paths = [];
-    for (const file of files) {
-      try {
-        // Re-encode to a standard JPEG/PNG before upload. Galleries on some
-        // phones (notably Android/MIUI) hand back a WebP/HEIF whose filename and
-        // MIME claim "image/jpeg", which passes the server's extension allowlist
-        // but fails its magic-byte check ("bytes do not match declared type").
-        // Decoding through the browser and re-encoding guarantees the bytes
-        // match the extension we send.
-        const normalized = await this._normalizeImageForUpload(file);
-        const path = await this._uploadPasteImage(sessionId, normalized);
-        paths.push(path);
-      } catch (err) {
-        this.showToast('Upload failed: ' + (err.message || 'unknown error'), 'error');
+    // Cap the batch and tell the user what got dropped (no silent truncation).
+    let capped = false;
+    if (files.length > this._maxBatchImages) {
+      files = files.slice(0, this._maxBatchImages);
+      capped = true;
+    }
+
+    const total = files.length;
+    let done = 0;
+    let failed = 0;
+    const results = new Array(total); // preserve selection order for insertion
+    const progress = () =>
+      this.showToast(`Uploading ${Math.min(done + 1, total)}/${total} image${total > 1 ? 's' : ''}…`, 'info');
+    progress();
+
+    // Bounded-concurrency worker pool over the file list.
+    let next = 0;
+    const worker = async () => {
+      for (;;) {
+        const i = next++;
+        if (i >= total) return;
+        try {
+          // Re-encode to a standard JPEG/PNG (and downscale very large images)
+          // before upload. Galleries on some phones (notably Android/MIUI) hand
+          // back a WebP/HEIF whose filename and MIME claim "image/jpeg", which
+          // passes the server's extension allowlist but fails its magic-byte
+          // check. Decoding through the browser and re-encoding guarantees the
+          // bytes match the extension we send — and shrinks huge photos so they
+          // fit the upload limit and iOS's <canvas> area cap.
+          const normalized = await this._normalizeImageForUpload(files[i]);
+          results[i] = await this._uploadPasteImage(sessionId, normalized);
+        } catch (err) {
+          failed++;
+          console.warn('Image upload failed:', err);
+          results[i] = null;
+        } finally {
+          done++;
+          if (done < total) progress();
+        }
       }
+    };
+    await Promise.all(Array.from({ length: Math.min(this._uploadConcurrency, total) }, () => worker()));
+
+    const paths = results.filter(Boolean);
+    if (paths.length > 0) {
+      // Insert all paths in one shot, space-separated, in selection order.
+      await this.sendInput(paths.join(' '));
     }
 
-    if (paths.length > 0) {
-      const pathStr = paths.join(' ');
-      await this.sendInput(pathStr);
-      this.showToast(paths.length + ' image' + (paths.length > 1 ? 's' : '') + ' ready', 'success');
-    }
+    // Final status: successes, plus any failures / cap so nothing is silent.
+    const parts = [];
+    if (paths.length > 0) parts.push(`${paths.length} image${paths.length > 1 ? 's' : ''} ready`);
+    if (failed > 0) parts.push(`${failed} failed`);
+    if (capped) parts.push(`max ${this._maxBatchImages} per batch`);
+    const tone = paths.length > 0 ? (failed > 0 || capped ? 'info' : 'success') : 'error';
+    this.showToast(parts.join(' · ') || 'No images uploaded', tone);
   },
 
   async _uploadPasteImage(sessionId, file) {
@@ -176,12 +220,24 @@ Object.assign(CodemanApp.prototype, {
       const height = img.naturalHeight;
       if (!width || !height) return file;
 
+      // Downscale very large images. Two reasons: (1) iOS Safari refuses to
+      // render a <canvas> larger than ~16.7M px (it returns a blank/null
+      // blob), so a 48MP photo would otherwise fail to re-encode and fall back
+      // to the original — which then trips the server's magic-byte check for
+      // HEIF mislabeled as JPEG. (2) It keeps multi-photo uploads fast and well
+      // under the size limit. Cap the longest edge so area stays safely below
+      // the canvas limit while still uploading a large, high-quality image.
+      const MAX_EDGE = 4096;
+      const scale = Math.min(1, MAX_EDGE / Math.max(width, height));
+      const w = Math.max(1, Math.round(width * scale));
+      const h = Math.max(1, Math.round(height * scale));
+
       const canvas = document.createElement('canvas');
-      canvas.width = width;
-      canvas.height = height;
+      canvas.width = w;
+      canvas.height = h;
       const ctx = canvas.getContext('2d');
       if (!ctx) return file;
-      ctx.drawImage(img, 0, 0);
+      ctx.drawImage(img, 0, 0, w, h);
 
       const mime = toPng ? 'image/png' : 'image/jpeg';
       const blob = await new Promise((resolve) => canvas.toBlob(resolve, mime, 0.92));

src/web/public/keyboard-accessory.js

@@ -100,6 +100,7 @@ const KeyboardAccessoryBar = {
       <button class="accessory-btn" data-action="esc" title="Escape">Esc</button>
       <button class="accessory-btn" data-action="init" title="/init">/init</button>
       <button class="accessory-btn" data-action="clear" title="/clear">/clear</button>
+      <button class="accessory-btn" data-action="compact" title="/compact">/compact</button>
       <button class="accessory-btn accessory-btn-dismiss" data-action="dismiss" title="Dismiss keyboard">
         <svg width="22" height="22" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="3">
           <path d="M19 9l-7 7-7-7"/>

src/web/public/notification-manager.js

@@ -273,7 +273,7 @@ class NotificationManager {
       const readClass = n.read ? '' : ' unread';
       const countLabel = n.count > 1 ? `<span class="notif-item-count">&times;${n.count}</span>` : '';
       const sessionChip = n.sessionName ? `<span class="notif-item-session">${escapeHtml(n.sessionName)}</span>` : '';
-      return `<div class="notif-item ${urgencyClass}${readClass}" data-notif-id="${n.id}" data-session-id="${n.sessionId || ''}" onclick="app.notificationManager.clickNotification('${escapeHtml(n.id)}')">
+      return `<div class="notif-item ${urgencyClass}${readClass}" data-notif-id="${n.id}" data-session-id="${n.sessionId || ''}" onclick="app.notificationManager.clickNotification(${escapeHtml(JSON.stringify(n.id))})">
         <div class="notif-item-header">
           <span class="notif-item-title">${escapeHtml(n.title)}${countLabel}</span>
           <span class="notif-item-time">${this.relativeTime(n.timestamp)}</span>

src/web/public/orchestrator-panel.js

@@ -392,10 +392,10 @@ Object.assign(CodemanApp.prototype, {
     let actions = '';
     if (orchState === 'executing' || orchState === 'failed') {
       if (phase.status === 'pending') {
-        actions += `<button class="orch-phase-btn" onclick="app.orchestratorSkipPhase('${phase.id}')" title="Skip">skip</button>`;
+        actions += `<button class="orch-phase-btn" onclick="app.orchestratorSkipPhase(${escapeHtml(JSON.stringify(phase.id))})" title="Skip">skip</button>`;
       }
       if (phase.status === 'failed') {
-        actions += `<button class="orch-phase-btn" onclick="app.orchestratorRetryPhase('${phase.id}')" title="Retry">retry</button>`;
+        actions += `<button class="orch-phase-btn" onclick="app.orchestratorRetryPhase(${escapeHtml(JSON.stringify(phase.id))})" title="Retry">retry</button>`;
       }
     }
 

src/web/public/panels-ui.js

@@ -753,8 +753,8 @@ Object.assign(CodemanApp.prototype, {
       const agentIcon = teammateInfo ? `<span class="subagent-icon teammate-dot teammate-color-${teammateInfo.color}">●</span>` : '<span class="subagent-icon">🤖</span>';
       html.push(`
         <div class="subagent-item ${statusClass} ${isActive ? 'selected' : ''}${teammateInfo ? ' is-teammate' : ''}"
-             onclick="app.selectSubagent('${escapeHtml(agent.agentId)}')"
-             ondblclick="app.openSubagentWindow('${escapeHtml(agent.agentId)}')"
+             onclick="app.selectSubagent(${escapeHtml(JSON.stringify(agent.agentId))})"
+             ondblclick="app.openSubagentWindow(${escapeHtml(JSON.stringify(agent.agentId))})"
              title="Double-click to open tracking window">
           <div class="subagent-header">
             ${agentIcon}
@@ -762,8 +762,8 @@ Object.assign(CodemanApp.prototype, {
             ${teammateBadge}
             ${modelBadge}
             <span class="subagent-status ${statusClass}">${agent.status}</span>
-            ${canKill ? `<button class="subagent-kill-btn" onclick="event.stopPropagation(); app.killSubagent('${escapeHtml(agent.agentId)}')" title="Kill agent">&#x2715;</button>` : ''}
-            <button class="subagent-window-btn" onclick="event.stopPropagation(); app.${hasWindow ? 'closeSubagentWindow' : 'openSubagentWindow'}('${escapeHtml(agent.agentId)}')" title="${hasWindow ? 'Close window' : 'Open in window'}">
+            ${canKill ? `<button class="subagent-kill-btn" onclick="event.stopPropagation(); app.killSubagent(${escapeHtml(JSON.stringify(agent.agentId))})" title="Kill agent">&#x2715;</button>` : ''}
+            <button class="subagent-window-btn" onclick="event.stopPropagation(); app.${hasWindow ? 'closeSubagentWindow' : 'openSubagentWindow'}(${escapeHtml(JSON.stringify(agent.agentId))})" title="${hasWindow ? 'Close window' : 'Open in window'}">
               ${hasWindow ? '✕' : '⧉'}
             </button>
           </div>
@@ -810,7 +810,7 @@ Object.assign(CodemanApp.prototype, {
           <span class="icon">${this.getToolIcon(a.tool)}</span>
           <span class="name">${escapeHtml(a.tool)}</span>
           <span class="detail">${escapeHtml(toolDetail.primary)}</span>
-          ${toolDetail.hasMore ? `<button class="tool-expand-btn" onclick="app.toggleToolParams('${escapeHtml(a.toolUseId)}')">▶</button>` : ''}
+          ${toolDetail.hasMore ? `<button class="tool-expand-btn" onclick="app.toggleToolParams(${escapeHtml(JSON.stringify(a.toolUseId))})">▶</button>` : ''}
           ${toolDetail.hasMore ? `<div class="tool-params-expanded" id="tool-params-${escapeHtml(a.toolUseId)}" style="display:none;"><pre>${escapeHtml(JSON.stringify(a.fullInput || a.input, null, 2))}</pre></div>` : ''}
         </div>`;
       } else if (a.type === 'tool_result') {
@@ -859,7 +859,7 @@ Object.assign(CodemanApp.prototype, {
         <span class="subagent-id" title="${escapeHtml(agent.description || agent.agentId)}">${escapeHtml(detailTitle.length > 60 ? detailTitle.substring(0, 60) + '...' : detailTitle)}</span>
         ${modelBadge}
         <span class="subagent-status ${agent.status}">${agent.status}</span>
-        <button class="subagent-transcript-btn" onclick="app.viewSubagentTranscript('${escapeHtml(agent.agentId)}')">
+        <button class="subagent-transcript-btn" onclick="app.viewSubagentTranscript(${escapeHtml(JSON.stringify(agent.agentId))})">
           View Full Transcript
         </button>
       </div>
@@ -1195,7 +1195,7 @@ Object.assign(CodemanApp.prototype, {
       parentDiv.dataset.parentSession = parentSessionId;
       parentDiv.innerHTML = `
         <span class="parent-label">from</span>
-        <span class="parent-name" onclick="app.selectSession('${escapeHtml(parentSessionId)}')">${escapeHtml(parentName)}</span>
+        <span class="parent-name" onclick="app.selectSession(${escapeHtml(JSON.stringify(parentSessionId))})">${escapeHtml(parentName)}</span>
       `;
       header.insertAdjacentElement('afterend', parentDiv);
     }
@@ -1687,7 +1687,7 @@ Object.assign(CodemanApp.prototype, {
           <span class="status running">terminal</span>
         </div>
         <div class="subagent-window-actions">
-          <button onclick="app.closeSubagentWindow('${escapeHtml(windowId)}')" title="Minimize to tab">─</button>
+          <button onclick="app.closeSubagentWindow(${escapeHtml(JSON.stringify(windowId))})" title="Minimize to tab">─</button>
         </div>
       </div>
       <div class="subagent-window-body teammate-terminal-body" id="subagent-window-body-${windowId}">
@@ -2200,7 +2200,7 @@ Object.assign(CodemanApp.prototype, {
         const fileName = path.split('/').pop();
         html.push(`
             <span class="project-insight-filepath"
-                  onclick="app.openLogViewerWindow('${escapeHtml(path)}', '${escapeHtml(tool.sessionId)}')"
+                  onclick="app.openLogViewerWindow(${escapeHtml(JSON.stringify(path))}, ${escapeHtml(JSON.stringify(tool.sessionId))})"
                   title="${escapeHtml(path)}">${escapeHtml(fileName)}</span>
         `);
       }
@@ -3099,7 +3099,7 @@ Object.assign(CodemanApp.prototype, {
           <span class="status streaming">streaming</span>
         </div>
         <div class="log-viewer-window-actions">
-          <button onclick="app.closeLogViewerWindow('${escapeHtml(windowId)}')" title="Close">×</button>
+          <button onclick="app.closeLogViewerWindow(${escapeHtml(JSON.stringify(windowId))})" title="Close">×</button>
         </div>
       </div>
       <div class="log-viewer-window-body" id="log-viewer-body-${windowId}">
@@ -3275,14 +3275,14 @@ Object.assign(CodemanApp.prototype, {
           <span class="size-badge">${sizeKB} KB</span>
         </div>
         <div class="image-popup-actions">
-          <button onclick="app.openImageInNewTab('${escapeHtml(imageUrl)}')" title="Open in new tab">↗</button>
-          <button onclick="app.closeImagePopup('${escapeHtml(imageId)}')" title="Close">×</button>
+          <button onclick="app.openImageInNewTab(${escapeHtml(JSON.stringify(imageUrl))})" title="Open in new tab">↗</button>
+          <button onclick="app.closeImagePopup(${escapeHtml(JSON.stringify(imageId))})" title="Close">×</button>
         </div>
       </div>
       <div class="image-popup-body">
         <img src="${imageUrl}" alt="${escapeHtml(fileName)}"
              onerror="this.parentElement.innerHTML='<div class=\\'image-error\\'>Failed to load image</div>'"
-             onclick="app.openImageInNewTab('${escapeHtml(imageUrl)}')" />
+             onclick="app.openImageInNewTab(${escapeHtml(JSON.stringify(imageUrl))})" />
       </div>
     `;
 
@@ -3505,9 +3505,9 @@ Object.assign(CodemanApp.prototype, {
         modelHtml = `<span class="monitor-model-badge ${modelShort}">${modelShort}</span>`;
       }
 
-      const sid = escapeHtml(muxSession.sessionId);
+      const sid = escapeHtml(JSON.stringify(muxSession.sessionId));
       html += `
-        <div class="process-item process-item-clickable" onclick="app.selectSession('${sid}')" title="Switch to session">
+        <div class="process-item process-item-clickable" onclick="app.selectSession(${sid})" title="Switch to session">
           <span class="monitor-status-badge ${statusClass}">${statusLabel}</span>
           <div class="process-info">
             <div class="process-name">${modelHtml} ${escapeHtml(muxSession.name || muxSession.muxName)}</div>
@@ -3520,7 +3520,7 @@ Object.assign(CodemanApp.prototype, {
             </div>
           </div>
           <div class="process-actions">
-            <button class="btn-toolbar btn-sm btn-danger" onclick="event.stopPropagation(); app.killMuxSession('${sid}')" title="Kill session">Kill</button>
+            <button class="btn-toolbar btn-sm btn-danger" onclick="event.stopPropagation(); app.killMuxSession(${sid})" title="Kill session">Kill</button>
           </div>
         </div>
       `;
@@ -3563,7 +3563,7 @@ Object.assign(CodemanApp.prototype, {
             </div>
           </div>
           <div class="process-actions">
-            ${agent.status !== 'completed' ? `<button class="btn-toolbar btn-sm btn-danger" onclick="app.killSubagent('${escapeHtml(agent.agentId)}')" title="Kill agent">Kill</button>` : ''}
+            ${agent.status !== 'completed' ? `<button class="btn-toolbar btn-sm btn-danger" onclick="app.killSubagent(${escapeHtml(JSON.stringify(agent.agentId))})" title="Kill agent">Kill</button>` : ''}
           </div>
         </div>
       `;

src/web/public/session-ui.js

@@ -1385,11 +1385,11 @@ Object.assign(CodemanApp.prototype, {
             <span class="case-manage-path">${escapeHtml(pathDisplay)}</span>
           </div>
           <div class="case-manage-actions">
-            <button class="case-manage-btn" onclick="app.moveCaseUp('${escapeHtml(c.name)}')"
+            <button class="case-manage-btn" onclick="app.moveCaseUp(${escapeHtml(JSON.stringify(c.name))})"
                     title="Move up" ${isFirst ? 'disabled' : ''}>&#x25B2;</button>
-            <button class="case-manage-btn" onclick="app.moveCaseDown('${escapeHtml(c.name)}')"
+            <button class="case-manage-btn" onclick="app.moveCaseDown(${escapeHtml(JSON.stringify(c.name))})"
                     title="Move down" ${isLast ? 'disabled' : ''}>&#x25BC;</button>
-            <button class="case-manage-btn case-manage-btn-delete" onclick="app.deleteCase('${escapeHtml(c.name)}')"
+            <button class="case-manage-btn case-manage-btn-delete" onclick="app.deleteCase(${escapeHtml(JSON.stringify(c.name))})"
                     title="Delete case">&#x2715;</button>
           </div>
         </div>
@@ -1484,14 +1484,14 @@ Object.assign(CodemanApp.prototype, {
       const isSelected = c.name === currentCase;
       html += `
         <button class="mobile-case-item ${isSelected ? 'selected' : ''}"
-                onclick="app.selectMobileCase('${escapeHtml(c.name)}')">
+                onclick="app.selectMobileCase(${escapeHtml(JSON.stringify(c.name))})">
           <span class="mobile-case-item-icon">
             <svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
               <path d="M22 19a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2V5a2 2 0 0 1 2-2h5l2 3h9a2 2 0 0 1 2 2z"/>
             </svg>
           </span>
           <span class="mobile-case-item-name">${escapeHtml(c.name)}</span>
-          <span class="mobile-case-item-delete" onclick="event.stopPropagation(); app.deleteCaseMobile('${escapeHtml(c.name)}')" title="Delete">
+          <span class="mobile-case-item-delete" onclick="event.stopPropagation(); app.deleteCaseMobile(${escapeHtml(JSON.stringify(c.name))})" title="Delete">
             <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
               <line x1="18" y1="6" x2="6" y2="18"/><line x1="6" y1="6" x2="18" y2="18"/>
             </svg>

src/web/public/styles.css

@@ -5573,6 +5573,45 @@ kbd {
   letter-spacing: 0.03em;
 }
 
+/* Ultracode run minimized-to-tab badge — same chip shape as the subagent badge but a
+   distinct purple (matches the ultracode window accent), and a SEPARATE class so the
+   incremental tab-update path (which keys on .tab-subagent-badge) never touches it.
+   Reuses the shared .subagent-dropdown for its restore/dismiss list. */
+.session-tab .tab-ultracode-badge {
+  position: relative;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 2px;
+  height: 16px;
+  padding: 0 5px;
+  border-radius: 8px;
+  font-size: 0.6rem;
+  cursor: pointer;
+  background: #a855f7;
+  color: white;
+  margin-left: 4px;
+  transition: transform 0.1s, background 0.15s;
+}
+@media (hover: hover) {
+  .session-tab .tab-ultracode-badge:hover {
+    background: #9333ea;
+    transform: scale(1.05);
+  }
+}
+.session-tab .tab-ultracode-badge .subagent-label {
+  font-size: 0.55rem;
+  font-weight: 600;
+  letter-spacing: 0.03em;
+}
+/* Per-item type glyph in the ULTRA dropdown (🧬 run vs 📄 transcript). */
+.subagent-dropdown-item .ultracode-dd-icon {
+  font-size: 0.7rem;
+  line-height: 1;
+  flex-shrink: 0;
+  margin-right: 1px;
+}
+
 /* Subagent Dropdown - compact, hover-triggered */
 .subagent-dropdown {
   display: none;
@@ -8647,13 +8686,6 @@ kbd {
   transform: scale(0.92);
   opacity: 0;
 }
-.ultracode-window.collapsed {
-  height: auto !important;
-  resize: none;
-}
-.ultracode-window.collapsed .ultracode-window-body {
-  display: none;
-}
 .ultracode-window-header {
   display: flex;
   align-items: center;
@@ -8744,6 +8776,32 @@ kbd {
   50% { opacity: 1; }
 }
 
+/* Agent-transcript window: spawned from an agent card (in a run window or the dock
+   panel), tied to its parent run window by its own connector line. Reuses the
+   .ultracode-window chrome; wider with a monospace transcript body. */
+.ultracode-agent-window {
+  width: 470px;
+  height: 420px;
+  border-color: #38bdf8;
+  box-shadow: 0 8px 32px rgba(0, 0, 0, 0.45), 0 0 0 1px rgba(56, 189, 248, 0.18);
+}
+.ultracode-agent-window .ultracode-window-body .uw-transcript {
+  margin: 0;
+  white-space: pre-wrap;
+  word-break: break-word;
+  font-family: var(--font-mono);
+  font-size: 0.68rem;
+  line-height: 1.45;
+  color: var(--text);
+}
+/* Distinguish the parent→agent line from the tab→run line: solid cyan, no dashes.
+   Must follow .ultracode-connection (equal specificity → source order wins). */
+.connection-line.ultracode-agent-connection {
+  stroke: #38bdf8;
+  stroke-dasharray: none;
+  filter: drop-shadow(0 0 2px rgba(0, 0, 0, 0.8)) drop-shadow(0 0 5px rgba(56, 189, 248, 0.85));
+}
+
 /* Plan-usage chip (App Settings → Display → "Plan Usage Limits"). Shows the
    live 5-hour + weekly plan limits parsed from the Claude statusline. Hidden by
    default via the marker class below; the server strips it at render when the

src/web/public/subagent-windows.js

@@ -33,10 +33,10 @@ Object.assign(CodemanApp.prototype, {
       const truncatedName = displayName.length > 25 ? displayName.substring(0, 25) + '…' : displayName;
       const statusClass = agent?.status || 'idle';
       agentItems.push(`
-        <div class="subagent-dropdown-item" onclick="event.stopPropagation(); app.restoreMinimizedSubagent('${escapeHtml(agentId)}', '${escapeHtml(sessionId)}')" title="Click to restore">
+        <div class="subagent-dropdown-item" onclick="event.stopPropagation(); app.restoreMinimizedSubagent(${escapeHtml(JSON.stringify(agentId))}, ${escapeHtml(JSON.stringify(sessionId))})" title="Click to restore">
           <span class="subagent-dropdown-status ${statusClass}"></span>
           <span class="subagent-dropdown-name">${escapeHtml(truncatedName)}</span>
-          <span class="subagent-dropdown-close" onclick="event.stopPropagation(); app.permanentlyCloseMinimizedSubagent('${escapeHtml(agentId)}', '${escapeHtml(sessionId)}')" title="Dismiss">&times;</span>
+          <span class="subagent-dropdown-close" onclick="event.stopPropagation(); app.permanentlyCloseMinimizedSubagent(${escapeHtml(JSON.stringify(agentId))}, ${escapeHtml(JSON.stringify(sessionId))})" title="Dismiss">&times;</span>
         </div>
       `);
     }
@@ -461,6 +461,10 @@ Object.assign(CodemanApp.prototype, {
     if (typeof this._appendUltracodeConnectionLines === 'function') {
       this._appendUltracodeConnectionLines(svg, rects);
     }
+    // Agent-transcript windows → their run window / tab (ultracode-windows.js).
+    if (typeof this._appendUltracodeAgentConnectionLines === 'function') {
+      this._appendUltracodeAgentConnectionLines(svg, rects);
+    }
   },
 
   // ═══════════════════════════════════════════════════════════════
@@ -695,7 +699,7 @@ Object.assign(CodemanApp.prototype, {
       parentSessionId && parentSessionName
         ? `<div class="subagent-window-parent" data-parent-session="${parentSessionId}">
           <span class="parent-label">from</span>
-          <span class="parent-name" onclick="app.selectSession('${escapeHtml(parentSessionId)}')">${escapeHtml(parentSessionName)}</span>
+          <span class="parent-name" onclick="app.selectSession(${escapeHtml(JSON.stringify(parentSessionId))})">${escapeHtml(parentSessionName)}</span>
         </div>`
         : '';
 
@@ -716,7 +720,7 @@ Object.assign(CodemanApp.prototype, {
           <span class="status ${agent.status}">${agent.status}</span>
         </div>
         <div class="subagent-window-actions">
-          <button onclick="app.closeSubagentWindow('${escapeHtml(agentId)}')" title="Minimize to tab">─</button>
+          <button onclick="app.closeSubagentWindow(${escapeHtml(JSON.stringify(agentId))})" title="Minimize to tab">─</button>
         </div>
       </div>
       ${parentHeader}

src/web/public/terminal-ui.js

@@ -2195,12 +2195,11 @@ Object.assign(CodemanApp.prototype, {
    * @returns {Promise<void>}
    */
   async sendInput(input) {
-    if (!this.activeSessionId) return;
-    await fetch(`/api/sessions/${this.activeSessionId}/input`, {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ input, useMux: true }),
-    });
+    if (!this.activeSessionId || !input) return;
+    // Route through the durable, exactly-once delivery layer (useMux for the
+    // POST fallback) so voice / keyboard-accessory / paste input also survives a
+    // dropped link instead of being lost in a single best-effort fetch.
+    this._sendInputAsync(this.activeSessionId, input, { useMux: true });
   },
 
   // ═══════════════════════════════════════════════════════════════

src/web/public/ultracode-panel.js

@@ -131,12 +131,15 @@ Object.assign(CodemanApp.prototype, {
     }
   },
 
-  // Phase 4: open an agent's live transcript by agentId. The workflow agent's
+  // Phase 4: fetch an agent's live transcript by agentId. The workflow agent's
   // agentId is byte-identical to the agent-<id>.jsonl stem already tracked by
   // subagent-watcher, so we reuse the existing transcript route — no watcher edits.
-  // Graceful when the agent isn't tracked yet / aged out / tracking disabled.
-  async openWorkflowAgentTranscript(agentId) {
-    if (!agentId) return;
+  // Returns { formatted: string[], entryCount } or null when nothing is available
+  // (queued / aged out of tracking / tracking disabled). Rendering into a connected
+  // in-page floating window lives in ultracode-windows.js (openUltracodeAgentWindow) —
+  // we no longer spawn a detached browser popup.
+  async _fetchWorkflowAgentTranscript(agentId) {
+    if (!agentId) return null;
     let data = null;
     try {
       const res = await fetch(`/api/subagents/${encodeURIComponent(agentId)}/transcript?format=formatted`);
@@ -147,21 +150,8 @@ Object.assign(CodemanApp.prototype, {
     const ok = data && data.success && data.data;
     const formatted = ok ? data.data.formatted : null;
     const entryCount = ok ? data.data.entryCount || 0 : 0;
-    if (!formatted || !entryCount) {
-      alert(
-        'No transcript available for this agent yet — it may be queued, aged out of tracking, or subagent tracking is disabled.'
-      );
-      return;
-    }
-    const win = window.open('', '_blank', 'width=860,height=640');
-    if (!win) return; // popup blocked
-    win.document.write(
-      `<html><head><title>Workflow agent ${escapeHtml(agentId)} transcript</title>` +
-        `<style>body{background:#1a1a2e;color:#eee;font-family:monospace;padding:20px}pre{white-space:pre-wrap;word-wrap:break-word}</style>` +
-        `</head><body><h2>Workflow agent ${escapeHtml(agentId)} (${entryCount} entries)</h2>` +
-        `<pre>${escapeHtml(formatted.join('\n'))}</pre></body></html>`
-    );
-    win.document.close();
+    if (!formatted || !entryCount) return null;
+    return { formatted, entryCount };
   },
 
   // ----- Render (debounced) -----
@@ -214,7 +204,7 @@ Object.assign(CodemanApp.prototype, {
       phasesHtml = `<div class="ultracode-phase-list">${chips.join('')}</div>`;
     }
     return (
-      `<div class="ultracode-run-item${active ? ' selected' : ''}" onclick="app.selectWorkflowRun('${escapeHtml(r.runId)}')">` +
+      `<div class="ultracode-run-item${active ? ' selected' : ''}" onclick="app.selectWorkflowRun(${escapeHtml(JSON.stringify(r.runId))})">` +
       `<div class="ultracode-run-head"><span class="ultracode-run-name">${name}</span>` +
       `<span class="ultracode-status ${statusCls}">${escapeHtml(status || '—')}</span></div>` +
       `<div class="ultracode-run-stats">${escapeHtml(stats)}</div>` +
@@ -262,13 +252,13 @@ Object.assign(CodemanApp.prototype, {
         const header =
           `<div class="ultracode-phase-header"><span>${escapeHtml(title)}</span>` +
           `<span class="ultracode-phase-sub">${this._fmtNum(tok)} tok · ${tools} tools</span></div>`;
-        return header + group.map((a) => this._workflowAgentCardHtml(a)).join('');
+        return header + group.map((a) => this._workflowAgentCardHtml(a, runId)).join('');
       })
       .join('');
     detail.innerHTML = html;
   },
 
-  _workflowAgentCardHtml(a) {
+  _workflowAgentCardHtml(a, runId) {
     const state = String(a.state || 'start');
     const stateCls = this._workflowAgentStateClass(state);
     const stateLabel = state === 'start' ? 'queued' : state === 'progress' ? 'running' : state;
@@ -289,7 +279,7 @@ Object.assign(CodemanApp.prototype, {
     const cardStateCls = state === 'done' ? ' uw-state-done' : state === 'progress' ? ' uw-state-working' : '';
     const cardAttrs = clickable
       ? ` class="ultracode-agent-card ultracode-agent-card--clickable${cardStateCls}" role="button" tabindex="0"` +
-        ` title="View transcript" onclick="app.openWorkflowAgentTranscript('${escapeHtml(a.agentId)}')"`
+        ` title="View transcript" onclick="app.openUltracodeAgentWindow(${escapeHtml(JSON.stringify(a.agentId))},${escapeHtml(JSON.stringify(runId || ''))})"`
       : ` class="ultracode-agent-card${cardStateCls}"`;
     return (
       `<div${cardAttrs}>` +

src/web/public/ultracode-windows.js

@@ -33,9 +33,12 @@
 Object.assign(CodemanApp.prototype, {
   /** Lazily seed the floating-window state maps (constructor also seeds them). */
   _ensureUltracodeWindowState() {
-    if (!this.ultracodeWindows) this.ultracodeWindows = new Map(); // runId -> { element, parentSessionId, dragListeners, collapsed }
+    if (!this.ultracodeWindows) this.ultracodeWindows = new Map(); // runId -> { element, parentSessionId, dragListeners }
     if (!this.ultracodeWindowsClosed) this.ultracodeWindowsClosed = new Set(); // runIds the user dismissed
     if (!this.ultracodeWindowCloseTimers) this.ultracodeWindowCloseTimers = new Map(); // runId -> setTimeout id
+    if (!this.ultracodeAgentWindows) this.ultracodeAgentWindows = new Map(); // agentId -> { element, runId, dragListeners }
+    if (!this.minimizedUltracodeRuns) this.minimizedUltracodeRuns = new Map(); // sessionId -> Set<runId> minimized to a tab
+    if (!this.minimizedUltracodeAgents) this.minimizedUltracodeAgents = new Map(); // sessionId -> Map<agentId,{runId,label}>
     if (this.ultracodeWindowZIndex === undefined) this.ultracodeWindowZIndex = 1000;
   },
 
@@ -86,6 +89,21 @@ Object.assign(CodemanApp.prototype, {
 
     const active = this._isWorkflowRunActive(run);
 
+    // Minimized to a tab — keep it there (don't re-pop a window). Clear the tab badge a
+    // short while after the run finishes, mirroring the floating window's finish grace.
+    if (this._isUltracodeRunMinimized(runId)) {
+      if (!active && !this.ultracodeWindowCloseTimers.has(runId)) {
+        const timer = setTimeout(() => {
+          this.ultracodeWindowCloseTimers.delete(runId);
+          this._removeMinimizedUltracodeRun(runId);
+          this.renderSessionTabs();
+          this.updateConnectionLines();
+        }, 8000);
+        this.ultracodeWindowCloseTimers.set(runId, timer);
+      }
+      return;
+    }
+
     if (active) {
       // Run is alive — cancel any pending auto-close.
       const pending = this.ultracodeWindowCloseTimers.get(runId);
@@ -134,11 +152,11 @@ Object.assign(CodemanApp.prototype, {
     const run = this.workflowRuns && this.workflowRuns.get(runId);
     if (!run) return;
     this.ultracodeWindowsClosed.delete(runId); // an explicit open overrides a past dismissal
+    this._removeMinimizedUltracodeRun(runId); // …and a past minimize-to-tab
     const existing = this.ultracodeWindows.get(runId);
     if (existing) {
-      // Already open — bring to front, expand if collapsed, refresh.
+      // Already open — bring to front and refresh.
       existing.element.style.zIndex = ++this.ultracodeWindowZIndex;
-      if (existing.collapsed) this.toggleUltracodeWindowCollapse(runId);
       this.renderUltracodeWindowContent(runId);
       this._fetchWorkflowRunDetail(runId);
       this.updateConnectionLines();
@@ -167,7 +185,7 @@ Object.assign(CodemanApp.prototype, {
           <span class="uw-status"></span>
         </div>
         <div class="ultracode-window-actions">
-          <button class="uw-min" type="button" title="Collapse">─</button>
+          <button class="uw-min" type="button" title="Minimize to tab">─</button>
           <button class="uw-close" type="button" title="Close">&times;</button>
         </div>
       </div>
@@ -198,7 +216,7 @@ Object.assign(CodemanApp.prototype, {
 
     win.querySelector('.uw-min').addEventListener('click', (e) => {
       e.stopPropagation();
-      this.toggleUltracodeWindowCollapse(runId);
+      this.minimizeUltracodeWindowToTab(runId);
     });
     win.querySelector('.uw-close').addEventListener('click', (e) => {
       e.stopPropagation();
@@ -211,19 +229,261 @@ Object.assign(CodemanApp.prototype, {
       nameEl.addEventListener('click', () => this.selectSession(parentSessionId));
     }
 
-    this.ultracodeWindows.set(runId, { element: win, parentSessionId, dragListeners, collapsed: false });
+    this.ultracodeWindows.set(runId, { element: win, parentSessionId, dragListeners });
 
     this.renderUltracodeWindowContent(runId);
     this._fetchWorkflowRunDetail(runId); // pull agents[] for the body
     this.updateConnectionLines();
   },
 
-  /** Collapse/expand the window to header-only (line stays connected). */
-  toggleUltracodeWindowCollapse(runId) {
+  // ── Minimize a run window into its originating tab (same idiom as subagent windows) ──
+
+  /** Is this run currently minimized to a tab (so auto-pop should leave it alone)? */
+  _isUltracodeRunMinimized(runId) {
+    if (!this.minimizedUltracodeRuns) return false;
+    for (const set of this.minimizedUltracodeRuns.values()) {
+      if (set.has(runId)) return true;
+    }
+    return false;
+  },
+
+  /** Drop a run from the minimized-to-tab tracking (all sessions, or a specific one). */
+  _removeMinimizedUltracodeRun(runId, sessionId) {
+    if (!this.minimizedUltracodeRuns) return;
+    if (sessionId) {
+      const set = this.minimizedUltracodeRuns.get(sessionId);
+      if (set) {
+        set.delete(runId);
+        if (!set.size) this.minimizedUltracodeRuns.delete(sessionId);
+      }
+      return;
+    }
+    for (const [sid, set] of this.minimizedUltracodeRuns) {
+      if (set.delete(runId) && !set.size) this.minimizedUltracodeRuns.delete(sid);
+    }
+  },
+
+  /**
+   * Minimize the floating run window into its originating session tab: record it as
+   * minimized (so a badge renders on the tab), genie-animate the window toward that
+   * tab, then remove the floating element. Restorable from the tab badge dropdown.
+   */
+  minimizeUltracodeWindowToTab(runId) {
+    this._ensureUltracodeWindowState();
     const data = this.ultracodeWindows.get(runId);
     if (!data) return;
-    data.collapsed = !data.collapsed;
-    data.element.classList.toggle('collapsed', data.collapsed);
+
+    let parentSessionId = data.parentSessionId;
+    if (!parentSessionId) {
+      const summary = this.workflowRuns && this.workflowRuns.get(runId);
+      parentSessionId = summary ? this._resolveUltracodeParentSession(summary) : null;
+    }
+    // No tab to fly into → fall back to a plain close so the window isn't orphaned.
+    if (!parentSessionId) {
+      this.closeUltracodeWindow(runId, true);
+      return;
+    }
+
+    // Cancel any pending finish auto-close — the badge owns the run's lifecycle now.
+    const pending = this.ultracodeWindowCloseTimers.get(runId);
+    if (pending) {
+      clearTimeout(pending);
+      this.ultracodeWindowCloseTimers.delete(runId);
+    }
+
+    if (!this.minimizedUltracodeRuns.has(parentSessionId)) this.minimizedUltracodeRuns.set(parentSessionId, new Set());
+    this.minimizedUltracodeRuns.get(parentSessionId).add(runId);
+
+    const element = data.element;
+    const dragListeners = data.dragListeners;
+    this._animateUltracodeWindowToTab(element, parentSessionId, () => {
+      this._teardownUltracodeDrag(dragListeners);
+      if (element) element.remove();
+      this.ultracodeWindows.delete(runId);
+      // Full rebuild so the tab badge renders (the incremental path only knows subagent badges).
+      this._fullRenderSessionTabs();
+      this.updateConnectionLines();
+    });
+  },
+
+  /** Genie the window toward the center of its tab, then invoke `done` to tear it down. */
+  _animateUltracodeWindowToTab(element, sessionId, done) {
+    const tab = sessionId ? document.querySelector(`.session-tab[data-id="${sessionId}"]`) : null;
+    if (!tab || !element) {
+      done();
+      return;
+    }
+    const w = element.getBoundingClientRect();
+    const t = tab.getBoundingClientRect();
+    const dx = t.left + t.width / 2 - (w.left + w.width / 2);
+    const dy = t.top + t.height / 2 - (w.top + w.height / 2);
+    element.style.transformOrigin = 'center center';
+    element.style.transition = 'transform 0.26s cubic-bezier(0.4, 0, 0.2, 1), opacity 0.26s ease';
+    element.style.pointerEvents = 'none';
+    requestAnimationFrame(() => {
+      element.style.transform = `translate(${dx}px, ${dy}px) scale(0.06)`;
+      element.style.opacity = '0';
+    });
+    let finished = false;
+    const finish = () => {
+      if (finished) return;
+      finished = true;
+      done();
+    };
+    element.addEventListener('transitionend', finish, { once: true });
+    setTimeout(finish, 320); // fallback in case transitionend doesn't fire
+  },
+
+  /** Tab badge (with restore/dismiss dropdown) for runs minimized to this session's tab. */
+  renderUltracodeTabBadge(sessionId) {
+    this._ensureUltracodeWindowState();
+    const runSet = this.minimizedUltracodeRuns.get(sessionId);
+    const agentMap = this.minimizedUltracodeAgents.get(sessionId);
+    const total = (runSet ? runSet.size : 0) + (agentMap ? agentMap.size : 0);
+    if (total === 0) return '';
+
+    const trunc = (s) => (s.length > 25 ? s.slice(0, 25) + '…' : s);
+    const items = [];
+    // Minimized run windows (🧬) first…
+    if (runSet) {
+      for (const runId of runSet) {
+        const run = this.workflowRuns && this.workflowRuns.get(runId);
+        const name = run ? run.workflowName || run.summary || runId : runId;
+        const statusCls = this._workflowStatusClass(run ? String(run.status || '') : '');
+        items.push(
+          `<div class="subagent-dropdown-item" onclick="event.stopPropagation(); app.restoreUltracodeRunFromTab(${escapeHtml(JSON.stringify(runId))},${escapeHtml(JSON.stringify(sessionId))})" title="Click to restore run">` +
+            `<span class="subagent-dropdown-status ${statusCls}"></span>` +
+            `<span class="ultracode-dd-icon">🧬</span>` +
+            `<span class="subagent-dropdown-name">${escapeHtml(trunc(name))}</span>` +
+            `<span class="subagent-dropdown-close" onclick="event.stopPropagation(); app.dismissMinimizedUltracodeRun(${escapeHtml(JSON.stringify(runId))},${escapeHtml(JSON.stringify(sessionId))})" title="Dismiss">&times;</span>` +
+            `</div>`
+        );
+      }
+    }
+    // …then minimized agent transcripts (📄).
+    if (agentMap) {
+      for (const [agentId, entry] of agentMap) {
+        const name = (entry && entry.label) || agentId;
+        items.push(
+          `<div class="subagent-dropdown-item" onclick="event.stopPropagation(); app.restoreUltracodeAgentFromTab(${escapeHtml(JSON.stringify(agentId))},${escapeHtml(JSON.stringify(sessionId))})" title="Click to restore transcript">` +
+            `<span class="subagent-dropdown-status"></span>` +
+            `<span class="ultracode-dd-icon">📄</span>` +
+            `<span class="subagent-dropdown-name">${escapeHtml(trunc(name))}</span>` +
+            `<span class="subagent-dropdown-close" onclick="event.stopPropagation(); app.dismissMinimizedUltracodeAgent(${escapeHtml(JSON.stringify(agentId))},${escapeHtml(JSON.stringify(sessionId))})" title="Dismiss">&times;</span>` +
+            `</div>`
+        );
+      }
+    }
+    const label = total === 1 ? 'ULTRA' : `ULTRA (${total})`;
+    return (
+      `<span class="tab-ultracode-badge" onmouseenter="app.showSubagentDropdown(this)" onmouseleave="app.scheduleHideSubagentDropdown(this)" onclick="event.stopPropagation(); app.pinSubagentDropdown(this);">` +
+      `<span class="subagent-label">${label}</span>` +
+      `<div class="subagent-dropdown" onmouseenter="app.cancelHideSubagentDropdown()" onmouseleave="app.scheduleHideSubagentDropdown(this.parentElement)">${items.join('')}</div>` +
+      `</span>`
+    );
+  },
+
+  /** Restore a minimized run from its tab badge: re-open the floating window. */
+  restoreUltracodeRunFromTab(runId, sessionId) {
+    this._ensureUltracodeWindowState();
+    this._removeMinimizedUltracodeRun(runId, sessionId);
+    this._fullRenderSessionTabs();
+    this.openUltracodeWindowForRun(runId);
+  },
+
+  /** Dismiss a minimized run from its tab badge (don't re-pop it). */
+  dismissMinimizedUltracodeRun(runId, sessionId) {
+    this._ensureUltracodeWindowState();
+    this._removeMinimizedUltracodeRun(runId, sessionId);
+    this.ultracodeWindowsClosed.add(runId);
+    this._fullRenderSessionTabs();
+    this.updateConnectionLines();
+  },
+
+  // ── Minimize an agent transcript window into its tab (same idiom as run windows) ──
+
+  /** Is this agent transcript currently minimized to a tab? */
+  _isUltracodeAgentMinimized(agentId) {
+    if (!this.minimizedUltracodeAgents) return false;
+    for (const map of this.minimizedUltracodeAgents.values()) {
+      if (map.has(agentId)) return true;
+    }
+    return false;
+  },
+
+  /** Look up a minimized agent's {runId,label} entry (across sessions). */
+  _getMinimizedUltracodeAgent(agentId) {
+    if (!this.minimizedUltracodeAgents) return null;
+    for (const map of this.minimizedUltracodeAgents.values()) {
+      if (map.has(agentId)) return map.get(agentId);
+    }
+    return null;
+  },
+
+  /** Drop an agent from minimized tracking (all sessions, or a specific one). */
+  _removeMinimizedUltracodeAgent(agentId, sessionId) {
+    if (!this.minimizedUltracodeAgents) return;
+    if (sessionId) {
+      const map = this.minimizedUltracodeAgents.get(sessionId);
+      if (map) {
+        map.delete(agentId);
+        if (!map.size) this.minimizedUltracodeAgents.delete(sessionId);
+      }
+      return;
+    }
+    for (const [sid, map] of this.minimizedUltracodeAgents) {
+      if (map.delete(agentId) && !map.size) this.minimizedUltracodeAgents.delete(sid);
+    }
+  },
+
+  /** Minimize an agent transcript window into the run's originating session tab. */
+  minimizeUltracodeAgentWindowToTab(agentId) {
+    this._ensureUltracodeWindowState();
+    const info = this.ultracodeAgentWindows.get(agentId);
+    if (!info) return;
+    const runId = info.runId;
+    const summary = runId && this.workflowRuns ? this.workflowRuns.get(runId) : null;
+    let parentSessionId = summary ? this._resolveUltracodeParentSession(summary) : null;
+    if (!parentSessionId && this.activeSessionId && this.sessions && this.sessions.has(this.activeSessionId)) {
+      parentSessionId = this.activeSessionId;
+    }
+    // No tab to fly into → plain close rather than orphan it.
+    if (!parentSessionId) {
+      this.closeUltracodeAgentWindow(agentId);
+      return;
+    }
+    const labelEl = info.element.querySelector('.uw-name');
+    const label = labelEl ? labelEl.textContent : agentId;
+    if (!this.minimizedUltracodeAgents.has(parentSessionId))
+      this.minimizedUltracodeAgents.set(parentSessionId, new Map());
+    this.minimizedUltracodeAgents.get(parentSessionId).set(agentId, { runId, label });
+
+    const element = info.element;
+    const dragListeners = info.dragListeners;
+    this._animateUltracodeWindowToTab(element, parentSessionId, () => {
+      this._teardownUltracodeDrag(dragListeners);
+      if (element) element.remove();
+      this.ultracodeAgentWindows.delete(agentId);
+      this._fullRenderSessionTabs();
+      this.updateConnectionLines();
+    });
+  },
+
+  /** Restore a minimized agent transcript from its tab badge: re-open its window. */
+  restoreUltracodeAgentFromTab(agentId, sessionId) {
+    this._ensureUltracodeWindowState();
+    const entry = this._getMinimizedUltracodeAgent(agentId);
+    const runId = entry ? entry.runId : null;
+    this._removeMinimizedUltracodeAgent(agentId, sessionId);
+    this._fullRenderSessionTabs();
+    this.openUltracodeAgentWindow(agentId, runId);
+  },
+
+  /** Dismiss a minimized agent transcript from its tab badge. */
+  dismissMinimizedUltracodeAgent(agentId, sessionId) {
+    this._ensureUltracodeWindowState();
+    this._removeMinimizedUltracodeAgent(agentId, sessionId);
+    this._fullRenderSessionTabs();
     this.updateConnectionLines();
   },
 
@@ -260,19 +520,149 @@ Object.assign(CodemanApp.prototype, {
     }
   },
 
+  // ── Agent-transcript windows ────────────────────────────────────────────────
+  // Clicking an agent card (in a run window OR the dock panel) opens the agent's
+  // live transcript as its OWN in-page floating window, line-tied to its parent run
+  // window (or the run's session tab when that window is closed). Replaces the old
+  // detached `window.open` browser popup so the transcript stays inside the same
+  // draggable, connector-line floating-window system as the run windows.
+
+  /** Open (or focus) the floating transcript window for a workflow agent. */
+  async openUltracodeAgentWindow(agentId, runId) {
+    this._ensureUltracodeWindowState();
+    if (!agentId) return;
+    this._removeMinimizedUltracodeAgent(agentId); // an explicit open overrides a past minimize
+    const existing = this.ultracodeAgentWindows.get(agentId);
+    if (existing && existing.element) {
+      // Already open — bring to front and refresh transcript.
+      existing.element.style.zIndex = ++this.ultracodeWindowZIndex;
+      this.updateConnectionLines();
+    } else if (!this.createUltracodeAgentWindow(agentId, runId)) {
+      return;
+    }
+    // Body shows a loading state until the fetch lands (re-fetch on focus too, so a
+    // still-running agent's transcript grows as you re-click).
+    const data = this._fetchWorkflowAgentTranscript ? await this._fetchWorkflowAgentTranscript(agentId) : null;
+    this.renderUltracodeAgentWindowContent(agentId, data);
+  },
+
+  /** Build and mount the floating agent-transcript window shell near its parent. */
+  createUltracodeAgentWindow(agentId, runId) {
+    this._ensureUltracodeWindowState();
+    if (this.ultracodeAgentWindows.has(agentId)) return this.ultracodeAgentWindows.get(agentId).element;
+
+    const label = this._ultracodeAgentLabel(agentId, runId) || agentId;
+    const win = document.createElement('div');
+    win.className = 'ultracode-window ultracode-agent-window spawning';
+    win.id = `ultracode-agent-window-${agentId}`;
+    win.style.zIndex = ++this.ultracodeWindowZIndex;
+    win.innerHTML = `
+      <div class="ultracode-window-header">
+        <div class="ultracode-window-title" title="${escapeHtml(label)} — transcript">
+          <span class="icon">📄</span>
+          <span class="uw-name">${escapeHtml(label)}</span>
+        </div>
+        <div class="ultracode-window-actions">
+          <button class="uw-min" type="button" title="Minimize to tab">─</button>
+          <button class="uw-close" type="button" title="Close">&times;</button>
+        </div>
+      </div>
+      <div class="ultracode-window-body">
+        <div class="subagent-empty">Loading transcript…</div>
+      </div>
+    `;
+
+    // Position: offset from the parent run window if it's open, else cascade.
+    const parentWin = runId ? this.ultracodeWindows.get(runId) : null;
+    if (parentWin && parentWin.element) {
+      const r = parentWin.element.getBoundingClientRect();
+      win.style.left = `${Math.max(8, Math.min(r.left + 40, window.innerWidth - 472))}px`;
+      win.style.top = `${Math.max(8, Math.min(r.top + 40, window.innerHeight - 160))}px`;
+    } else {
+      const n = this.ultracodeAgentWindows.size;
+      win.style.left = `${Math.min(140 + n * 28, Math.max(8, window.innerWidth - 472))}px`;
+      win.style.top = `${110 + n * 28}px`;
+    }
+
+    document.body.appendChild(win);
+    requestAnimationFrame(() => win.classList.remove('spawning'));
+
+    const header = win.querySelector('.ultracode-window-header');
+    const dragListeners = this.makeWindowDraggable(win, header);
+    win.querySelector('.uw-min').addEventListener('click', (e) => {
+      e.stopPropagation();
+      this.minimizeUltracodeAgentWindowToTab(agentId);
+    });
+    win.querySelector('.uw-close').addEventListener('click', (e) => {
+      e.stopPropagation();
+      this.closeUltracodeAgentWindow(agentId);
+    });
+
+    this.ultracodeAgentWindows.set(agentId, { element: win, runId, dragListeners });
+    this.updateConnectionLines();
+    return win;
+  },
+
+  /** Resolve a human label for an agent from the run's fetched detail.agents[]. */
+  _ultracodeAgentLabel(agentId, runId) {
+    const detail = this.workflowRunDetails && runId ? this.workflowRunDetails.get(runId) : null;
+    const agents = detail && Array.isArray(detail.agents) ? detail.agents : null;
+    if (agents) {
+      const found = agents.find((a) => a.agentId === agentId);
+      if (found && found.label) return found.label;
+    }
+    return null;
+  },
+
+  /** Fill an agent window's body with the fetched transcript (or a friendly empty state). */
+  renderUltracodeAgentWindowContent(agentId, data) {
+    const info = this.ultracodeAgentWindows.get(agentId);
+    if (!info || !info.element) return;
+    const body = info.element.querySelector('.ultracode-window-body');
+    if (!body) return;
+    if (!data || !data.formatted || !data.entryCount) {
+      body.innerHTML =
+        '<div class="subagent-empty">No transcript available yet — the agent may be queued, aged out of tracking, or subagent tracking is disabled.</div>';
+      return;
+    }
+    const text = escapeHtml(data.formatted.join('\n'));
+    body.innerHTML = `<div class="uw-summary">${data.entryCount} entries</div><pre class="uw-transcript">${text}</pre>`;
+  },
+
+  /** Close one floating agent-transcript window. */
+  closeUltracodeAgentWindow(agentId) {
+    this._ensureUltracodeWindowState();
+    const info = this.ultracodeAgentWindows.get(agentId);
+    if (!info) return;
+    this._teardownUltracodeDrag(info.dragListeners);
+    if (info.element) info.element.remove();
+    this.ultracodeAgentWindows.delete(agentId);
+    this.updateConnectionLines();
+  },
+
   /** Tear down every floating window (called on SSE reconnect; keeps user dismissals). */
   removeAllUltracodeWindows() {
     this._ensureUltracodeWindowState();
-    const had = this.ultracodeWindows.size > 0;
+    const hadMinimized = this.minimizedUltracodeRuns.size > 0 || this.minimizedUltracodeAgents.size > 0;
+    const had = this.ultracodeWindows.size > 0 || this.ultracodeAgentWindows.size > 0;
     for (const [, data] of this.ultracodeWindows) {
       this._teardownUltracodeDrag(data.dragListeners);
       if (data.element) data.element.remove();
     }
     this.ultracodeWindows.clear();
+    for (const [, info] of this.ultracodeAgentWindows) {
+      this._teardownUltracodeDrag(info.dragListeners);
+      if (info.element) info.element.remove();
+    }
+    this.ultracodeAgentWindows.clear();
     for (const t of this.ultracodeWindowCloseTimers.values()) clearTimeout(t);
     this.ultracodeWindowCloseTimers.clear();
+    this.minimizedUltracodeRuns.clear();
+    this.minimizedUltracodeAgents.clear();
     // Redraw so the now-orphaned connector lines are cleared from the shared SVG.
     if (had) this.updateConnectionLines();
+    // Drop any now-stale tab badges.
+    if (hadMinimized) this.renderSessionTabs();
   },
 
   /** When the feature is toggled on, pop windows for any currently-active runs. */
@@ -355,7 +745,7 @@ Object.assign(CodemanApp.prototype, {
         const header =
           `<div class="ultracode-phase-header"><span>${escapeHtml(title)}</span>` +
           `<span class="ultracode-phase-sub">${this._fmtNum(tok)} tok · ${tools} tools</span></div>`;
-        return header + group.map((a) => this._workflowAgentCardHtml(a)).join('');
+        return header + group.map((a) => this._workflowAgentCardHtml(a, run.runId)).join('');
       })
       .join('');
     return head + grid;
@@ -408,4 +798,51 @@ Object.assign(CodemanApp.prototype, {
       svg.appendChild(line);
     }
   },
+
+  /**
+   * Append agent-window → parent connector lines into the shared SVG. Parent is the
+   * agent's run floating window when open, else the run's session tab. Called right
+   * after `_appendUltracodeConnectionLines` so it shares the same batched pass + the
+   * tab-rect cache.
+   */
+  _appendUltracodeAgentConnectionLines(svg, rects) {
+    this._ensureUltracodeWindowState();
+    if (!svg || !this.ultracodeAgentWindows.size) return;
+    if (!rects) rects = new Map();
+
+    for (const [agentId, info] of this.ultracodeAgentWindows) {
+      if (!info.element) continue;
+      const winRect = info.element.getBoundingClientRect();
+      // Anchor: parent run window bottom-center if open, else the run's tab.
+      let px, py;
+      const runWin = info.runId ? this.ultracodeWindows.get(info.runId) : null;
+      if (runWin && runWin.element) {
+        const pr = runWin.element.getBoundingClientRect();
+        px = pr.left + pr.width / 2;
+        py = pr.bottom;
+      } else {
+        const summary = info.runId && this.workflowRuns ? this.workflowRuns.get(info.runId) : null;
+        const parentSessionId = summary ? this._resolveUltracodeParentSession(summary) : null;
+        if (!parentSessionId) continue;
+        const tabKey = 'tab:' + parentSessionId;
+        if (!rects.has(tabKey)) {
+          const tab = document.querySelector(`.session-tab[data-id="${parentSessionId}"]`);
+          if (tab) rects.set(tabKey, tab.getBoundingClientRect());
+        }
+        const tabRect = rects.get(tabKey);
+        if (!tabRect) continue;
+        px = tabRect.left + tabRect.width / 2;
+        py = tabRect.bottom;
+      }
+      const x2 = winRect.left + winRect.width / 2;
+      const y2 = winRect.top;
+      const midY = (py + y2) / 2;
+      const path = `M ${px} ${py} C ${px} ${midY}, ${x2} ${midY}, ${x2} ${y2}`;
+      const line = document.createElementNS('http://www.w3.org/2000/svg', 'path');
+      line.setAttribute('d', path);
+      line.setAttribute('class', 'connection-line ultracode-connection ultracode-agent-connection');
+      line.setAttribute('data-agent-id', agentId);
+      svg.appendChild(line);
+    }
+  },
 });

src/web/routes/session-routes.ts

@@ -60,6 +60,7 @@ import { MAX_CONCURRENT_SESSIONS } from '../../config/map-limits.js';
 import { RunSummaryTracker } from '../../run-summary.js';
 
 import { MAX_INPUT_LENGTH, MAX_SESSION_NAME_LENGTH } from '../../config/terminal-limits.js';
+import { MAX_PASTE_IMAGE_BYTES } from '../../config/buffer-limits.js';
 import { dataPath } from '../../config/instance.js';
 
 // Path to linked-cases registry (same file used by case-routes resolveCasePath)
@@ -662,7 +663,7 @@ export function registerSessionRoutes(
 
   app.post('/api/sessions/:id/input', async (req) => {
     const { id } = req.params as { id: string };
-    const { input, useMux } = parseBody(SessionInputWithLimitSchema, req.body);
+    const { input, useMux, seq, clientId } = parseBody(SessionInputWithLimitSchema, req.body);
     const session = findSessionOrFail(ctx, id);
 
     const inputStr = String(input);
@@ -673,6 +674,13 @@ export function registerSessionRoutes(
       );
     }
 
+    // Reliable delivery (POST fallback when the WebSocket is down): a 2xx IS the
+    // client's ACK, so a tagged duplicate redelivery must still return 200 but
+    // skip the write. Untagged requests (curl/legacy) always apply.
+    if (typeof clientId === 'string' && typeof seq === 'number' && !session.shouldApplyInput(clientId, seq)) {
+      return {};
+    }
+
     // Write input to PTY. Direct write is synchronous; writeViaMux
     // (tmux send-keys) is fire-and-forget to avoid blocking the HTTP response.
     if (useMux) {
@@ -1703,7 +1711,7 @@ export function registerSessionRoutes(
   // ═══════════════════════════════════════════════════════════════
 
   const ALLOWED_IMAGE_EXTS = new Set(['.png', '.jpg', '.jpeg', '.gif', '.webp', '.bmp']);
-  // The 10MB size cap is enforced by @fastify/multipart (registered in server.ts).
+  // The per-file size cap (MAX_PASTE_IMAGE_BYTES) is enforced by @fastify/multipart (registered in server.ts).
 
   app.post('/api/sessions/:id/paste-image', async (req, reply) => {
     // CSRF defense: state-changing routes must come from same origin.
@@ -1740,7 +1748,7 @@ export function registerSessionRoutes(
     const { id } = req.params as { id: string };
 
     // Rate limit per (IP, sessionId): 30/min. Defends against disk-fill DoS
-    // — even an authenticated attacker can otherwise loop 10MB POSTs.
+    // — even an authenticated attacker can otherwise loop large image POSTs.
     if (!consumePasteToken(`${req.ip}:${id}`)) {
       reply.code(429);
       reply.header('Retry-After', '60');
@@ -1754,8 +1762,9 @@ export function registerSessionRoutes(
       return createErrorResponse(ApiErrorCode.INVALID_INPUT, 'Expected multipart/form-data');
     }
 
-    // Read the single file part. @fastify/multipart enforces the 10MB size cap
-    // and the 1-file/4-field count limits (server.ts), replacing a hand-rolled
+    // Read the single file part. @fastify/multipart enforces the per-file size
+    // cap (MAX_PASTE_IMAGE_BYTES) and the 1-file/4-field count limits (server.ts),
+    // replacing a hand-rolled
     // boundary scanner with several bugs: literal boundary matches anywhere in
     // body, LF-only clients silently corrupted the last byte (hard-coded \r\n
     // offsets), no part-count cap.
@@ -1779,7 +1788,8 @@ export function registerSessionRoutes(
       imageBytes = await part.toBuffer();
     } catch (err: unknown) {
       reply.code(413);
-      return createErrorResponse(ApiErrorCode.INVALID_INPUT, getErrorMessage(err) || 'File too large (max 10MB)');
+      const maxMb = Math.round(MAX_PASTE_IMAGE_BYTES / (1024 * 1024));
+      return createErrorResponse(ApiErrorCode.INVALID_INPUT, getErrorMessage(err) || `File too large (max ${maxMb}MB)`);
     }
     if (imageBytes.length === 0) {
       reply.code(400);
@@ -1843,9 +1853,22 @@ export function registerSessionRoutes(
       }
     } catch (err: unknown) {
       if ((err as NodeJS.ErrnoException).code !== 'ENOENT') throw err;
-      // Non-recursive mkdir: errors on EEXIST and does not follow symlinks for
-      // the leaf. session.workingDir is guaranteed to exist (live session).
-      await fs.mkdir(imageDir);
+      // Non-recursive mkdir: does not follow symlinks for the leaf.
+      // session.workingDir is guaranteed to exist (live session).
+      try {
+        await fs.mkdir(imageDir);
+      } catch (mkErr: unknown) {
+        // Concurrent uploads (a batch of photos) race to create .claude-images —
+        // the losers get EEXIST. Treat an already-present REAL directory as
+        // success, but re-verify it isn't a symlink a racing actor planted
+        // (preserve the symlink-safety guarantee above).
+        if ((mkErr as NodeJS.ErrnoException).code !== 'EEXIST') throw mkErr;
+        const raceStat = await fs.lstat(imageDir);
+        if (raceStat.isSymbolicLink() || !raceStat.isDirectory()) {
+          reply.code(403);
+          return createErrorResponse(ApiErrorCode.INVALID_INPUT, '.claude-images is not a regular directory');
+        }
+      }
     }
     // Date.now() collides on same-ms uploads from two tabs (last-write wins
     // silently). Append 8 hex chars so concurrent pastes get distinct names.

src/web/routes/ws-routes.ts

@@ -21,8 +21,11 @@
  *     {"t":"o","d":"..."} — terminal output
  *     {"t":"c"}           — clear terminal
  *     {"t":"r"}           — needs refresh (reload buffer)
+ *     {"t":"ia","seq":N}  — input ACK (echoes the seq of an applied/deduped input frame)
  *   Client -> Server:
- *     {"t":"i","d":"..."} — input (keystroke or paste)
+ *     {"t":"i","d":"...","seq":N,"cid":"..."} — input (keystroke or paste). seq+cid are
+ *                          optional reliable-delivery tags: the server applies each
+ *                          (cid,seq) at-most-once and ACKs with {"t":"ia","seq":N}.
  *     {"t":"z","c":N,"r":N,"f":bool} — resize terminal (f=true forces SIGWINCH even if dims unchanged)
  */
 
@@ -123,10 +126,22 @@ export function registerWsRoutes(app: FastifyInstance, ctx: SessionPort, getHost
         const msg = JSON.parse(String(raw));
         if (msg.t === 'i' && typeof msg.d === 'string') {
           if (msg.d.length > MAX_INPUT_LENGTH) return;
-          // Typed input from a claim-holding desktop keeps the claim "hot"
-          // and re-asserts the desktop layout after a mobile override.
-          if (holdsDesktopClaim) session.noteDesktopActivity();
-          session.write(msg.d);
+          // Reliable delivery: when the frame carries a clientId + seq, apply it
+          // exactly once (skip a duplicate redelivery) but ACK it regardless so
+          // the client can drop it from its durable queue. Frames without seq
+          // (legacy/other tools) are applied as-is — no behavior change.
+          const cid = typeof msg.cid === 'string' ? msg.cid : null;
+          const seq = Number.isInteger(msg.seq) ? (msg.seq as number) : null;
+          const apply = cid && seq !== null ? session.shouldApplyInput(cid, seq) : true;
+          if (apply) {
+            // Typed input from a claim-holding desktop keeps the claim "hot"
+            // and re-asserts the desktop layout after a mobile override.
+            if (holdsDesktopClaim) session.noteDesktopActivity();
+            session.write(msg.d);
+          }
+          if (seq !== null && socket.readyState === 1) {
+            socket.send(`{"t":"ia","seq":${seq}}`);
+          }
         } else if (
           msg.t === 'z' &&
           Number.isInteger(msg.c) &&

src/web/schemas.ts

@@ -469,6 +469,15 @@ export const SettingsUpdateSchema = z
 export const SessionInputWithLimitSchema = z.object({
   input: z.string().max(100000), // 100KB max input
   useMux: z.boolean().optional(),
+  // Reliable-delivery dedup (optional; absent for curl/legacy clients). The web
+  // client tags each input with a stable clientId + a monotonic per-session seq
+  // and redelivers anything it hasn't seen ACKed (e.g. a frame silently dropped
+  // by a half-open WebSocket on a flaky link). The server applies each (clientId,
+  // seq) at-most-once via Session.shouldApplyInput so a redelivery can't type the
+  // prompt twice. `.optional()` (not `.nullish()`) — the client omits them when
+  // unset rather than sending null. See docs/reliable-input-delivery.md.
+  seq: z.number().int().nonnegative().optional(),
+  clientId: z.string().max(128).optional(),
 });
 
 // ========== Session Mutation Routes ==========

src/web/server.ts

@@ -128,6 +128,7 @@ import {
 } from '../utils/index.js';
 import type { EventLoopMonitorHandle } from '../utils/index.js';
 import { MAX_CONCURRENT_SESSIONS, MAX_SSE_CLIENTS } from '../config/map-limits.js';
+import { MAX_PASTE_IMAGE_BYTES } from '../config/buffer-limits.js';
 import { SseEvent } from './sse-events.js';
 import { getLatestPlanUsage } from './plan-usage-latest.js';
 import type { ScheduledRun } from './ports/index.js';
@@ -679,8 +680,8 @@ export class WebServer extends EventEmitter {
     // last byte (hard-coded \r\n offsets), and there was no part-count cap.
     await this.app.register(fastifyMultipart, {
       limits: {
-        fileSize: 10 * 1024 * 1024, // 10MB per file
-        files: 1, // paste-image only ever sends one file
+        fileSize: MAX_PASTE_IMAGE_BYTES, // per file (default 50MB) — large phone photos / screenshots
+        files: 1, // paste-image sends one file per request (clients batch up to 20 requests)
         fields: 4, // small headroom for accompanying form fields
       },
     });

test/mocks/mock-session.ts

@@ -39,6 +39,16 @@ export class MockSession extends EventEmitter {
     return true;
   }
 
+  /** Exactly-once input dedup — mirrors Session.shouldApplyInput so route tests
+   *  exercising the reliable-delivery path behave like production. */
+  private _appliedInputSeq = new Map<string, number>();
+  shouldApplyInput(clientId: string, seq: number): boolean {
+    const last = this._appliedInputSeq.get(clientId);
+    if (last !== undefined && seq <= last) return false;
+    this._appliedInputSeq.set(clientId, seq);
+    return true;
+  }
+
   /** Get the last written data */
   get lastWrite(): string | undefined {
     return this.writeBuffer[this.writeBuffer.length - 1];

test/reliable-input-dedup.test.ts

@@ -0,0 +1,73 @@
+/**
+ * @fileoverview Exactly-once input delivery — Session.shouldApplyInput dedup.
+ *
+ * Guards the server half of the reliable-input-delivery feature: the web client
+ * tags each input frame with a stable clientId + a monotonic per-session seq and
+ * redelivers anything it hasn't seen ACKed (a half-open socket silently drops
+ * frames on a flaky link). shouldApplyInput must apply each (clientId, seq)
+ * exactly once so a redelivery can never type the prompt twice — while still
+ * applying untagged input (curl/legacy) unconditionally at the call sites.
+ *
+ * See docs/reliable-input-delivery.md.
+ */
+import { describe, it, expect } from 'vitest';
+import { Session } from '../src/session.js';
+
+function makeSession(): Session {
+  // workingDir is the only required field; no PTY is spawned until start(),
+  // and TmuxManager no-ops under VITEST — so this is a cheap, side-effect-free
+  // instance for exercising the pure dedup bookkeeping.
+  return new Session({ workingDir: '/tmp' });
+}
+
+describe('Session.shouldApplyInput (exactly-once input dedup)', () => {
+  it('applies a fresh (clientId, seq) exactly once', () => {
+    const s = makeSession();
+    expect(s.shouldApplyInput('clientA', 1)).toBe(true);
+    // Same seq redelivered (lost ACK) — must NOT apply again.
+    expect(s.shouldApplyInput('clientA', 1)).toBe(false);
+  });
+
+  it('applies strictly increasing seqs and rejects stale ones', () => {
+    const s = makeSession();
+    expect(s.shouldApplyInput('c', 1)).toBe(true);
+    expect(s.shouldApplyInput('c', 2)).toBe(true);
+    expect(s.shouldApplyInput('c', 3)).toBe(true);
+    // Out-of-order / replayed lower seqs are duplicates.
+    expect(s.shouldApplyInput('c', 2)).toBe(false);
+    expect(s.shouldApplyInput('c', 1)).toBe(false);
+    // The next genuinely-new seq still applies.
+    expect(s.shouldApplyInput('c', 4)).toBe(true);
+  });
+
+  it('tracks each client independently', () => {
+    const s = makeSession();
+    expect(s.shouldApplyInput('a', 5)).toBe(true);
+    // A different client at seq 1 is not shadowed by client a's higher seq.
+    expect(s.shouldApplyInput('b', 1)).toBe(true);
+    expect(s.shouldApplyInput('b', 1)).toBe(false);
+    expect(s.shouldApplyInput('a', 6)).toBe(true);
+  });
+
+  it('tolerates a seq gap (skips never collapse a new seq to a duplicate)', () => {
+    const s = makeSession();
+    expect(s.shouldApplyInput('c', 1)).toBe(true);
+    // Client jumped seq (e.g. resumed after a reload that kept the counter).
+    expect(s.shouldApplyInput('c', 100)).toBe(true);
+    expect(s.shouldApplyInput('c', 100)).toBe(false);
+    expect(s.shouldApplyInput('c', 50)).toBe(false);
+    expect(s.shouldApplyInput('c', 101)).toBe(true);
+  });
+
+  it('keeps recent clients dedup-correct past the eviction bound', () => {
+    const s = makeSession();
+    // Far exceed MAX_INPUT_DEDUP_CLIENTS (256) with one-shot clients, then prove
+    // a freshly-active client is still deduped correctly (MRU eviction).
+    for (let i = 0; i < 400; i++) {
+      expect(s.shouldApplyInput(`oneshot-${i}`, 1)).toBe(true);
+    }
+    expect(s.shouldApplyInput('recent', 1)).toBe(true);
+    expect(s.shouldApplyInput('recent', 1)).toBe(false);
+    expect(s.shouldApplyInput('recent', 2)).toBe(true);
+  });
+});

test/routes/session-routes.test.ts

@@ -355,6 +355,43 @@ describe('session-routes', () => {
       const body = JSON.parse(res.body);
       expect(body.success).toBe(false);
     });
+
+    it('applies a tagged (clientId, seq) input exactly once on redelivery', async () => {
+      const url = `/api/sessions/${harness.ctx._sessionId}/input`;
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const session = harness.ctx.sessions.get(harness.ctx._sessionId) as any;
+      session.writeBuffer.length = 0;
+
+      const post = (payload: unknown) => harness.app.inject({ method: 'POST', url, payload });
+
+      // First delivery of seq 1 — applied (200, written once).
+      const first = await post({ input: 'prompt', seq: 1, clientId: 'cid-1' });
+      expect(first.statusCode).toBe(200);
+
+      // Redelivery of the SAME seq (client never saw the ACK) — still 200, but
+      // must NOT write again.
+      const dup = await post({ input: 'prompt', seq: 1, clientId: 'cid-1' });
+      expect(dup.statusCode).toBe(200);
+
+      // A genuinely new seq — applied.
+      const next = await post({ input: '\r', seq: 2, clientId: 'cid-1' });
+      expect(next.statusCode).toBe(200);
+
+      expect(session.writeBuffer).toEqual(['prompt', '\r']);
+    });
+
+    it('always applies untagged input (curl/legacy, no dedup)', async () => {
+      const url = `/api/sessions/${harness.ctx._sessionId}/input`;
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const session = harness.ctx.sessions.get(harness.ctx._sessionId) as any;
+      session.writeBuffer.length = 0;
+
+      const post = () => harness.app.inject({ method: 'POST', url, payload: { input: 'x' } });
+      await post();
+      await post();
+      // No seq/clientId ⇒ no dedup ⇒ both writes land.
+      expect(session.writeBuffer).toEqual(['x', 'x']);
+    });
   });
 
   // ========== POST /api/sessions/:id/resize ==========