- Remove TUI from git (moved to .gitignore for local development)
- Remove React/Ink dependencies (unused in published version)
- Remove tui command from CLI
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Reduce getRecentSubagents window from 60 to 15 minutes
- Skip old/completed agents when restoring window states (10 min cutoff)
- Clear old activity data when new agent with same ID is discovered
- Force close existing window for agentId before opening fresh one
Fixes issue where old prompts/input from previous Ralph Loop runs
were displayed when clicking on subagent windows.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements the Execution Optimizer Rework plan that enables the system
to actually use the optimizer metadata (parallelGroups, agentType,
recommendedModel, requiresFreshContext, estimatedTokens) that was
previously generated but ignored.
New components:
- ExecutionBridge: Central coordinator that loads optimized plans,
manages parallel execution within groups, and coordinates with
SpawnOrchestrator for session-based execution
- ModelSelector: Routes tasks to appropriate models (opus/sonnet/haiku)
based on user defaults and agent type overrides. Optimizer
recommendations are advisory only - user preferences always win
- GroupScheduler: Builds topologically ordered execution groups,
manages dependencies, determines execution mode (session vs task-tool)
- ContextManager: Handles fresh context requirements via /clear+/init
or new session spawning
Features:
- Parallel task execution within groups (configurable limit)
- Group-level dependency tracking (lower groups complete first)
- Partial failure handling (continue with non-dependent tasks)
- Model configuration in App Settings > Models tab
- Agent type overrides (explore, implement, test, review)
- Execution control API endpoints (start, pause, resume, cancel)
- SSE events for real-time execution progress visibility
- Execution history tracking
API endpoints:
- GET/POST/PUT /api/execution/* for execution control
- GET/PUT /api/execution/model-config for model settings
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Session updates (like renaming) were broadcasting toDetailedState() which
doesn't include respawn controller state. The frontend would then overwrite
the session object, losing respawnEnabled/respawnConfig/respawn fields.
Added getSessionStateWithRespawn() helper that includes respawn controller
state in all session:updated broadcasts.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements image file watching in session working directories with
automatic popup display in the web UI:
- Add ImageWatcher class using chokidar for cross-platform file watching
- Add ImageDetectedEvent type for SSE broadcasting
- Integrate with session lifecycle (watch on create, unwatch on delete)
- Add image:detected SSE event handler in frontend
- Create draggable image popup window with open-in-new-tab support
- Add CSS styling with compact wizard layout
Supports: png, jpg, jpeg, gif, webp, bmp, svg
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds a new Phase 0 research agent that runs before all other analysis agents:
- Web search for GitHub repos, documentation, tutorials
- Codebase exploration for existing patterns
- Technical recommendations and potential challenges
- Creates enriched task description for downstream agents
- Research context injected into all analysis agents via {RESEARCH_CONTEXT}
- Saves output to research/ folder with prompt.md and result.json
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The plan generation API (/api/generate-plan-detailed) was incorrectly
detecting client disconnection because it listened to req.raw.on('close')
which fires when the HTTP request body finishes parsing, not when the
actual TCP connection closes.
This caused the error "Failed to parse plan - no JSON array found" because
the server would cancel all subagent sessions almost immediately after
starting them.
Fix:
- Changed from req.raw.on('close') to socket.on('close')
- Added responseSent flag to only cancel if response hasn't been sent
- Added E2E test to verify the fix
Tested with:
- Simple plan generation: 61 items, 138.5s, quality 0.82
- Smartphone app plan: 53 items, 93.4s, quality 0.75
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- SSE event listener cleanup: Track all listeners and remove on reconnect
to prevent listener accumulation (60+ listeners per connect cycle)
- Floating window cleanup: Remove drag listeners, ResizeObservers on close
- Session creation mutex: Prevent race conditions from concurrent requests
- Screen kill verification: Confirm processes are dead after kill signals
- State store circuit breaker: Prevent cascading failures with backup/recovery
- PTY spawn error handling: Wrap all spawn calls with proper error emission
- SSE graceful shutdown: Notify clients before server stops
- Async file writes: Replace blocking writeFileSync in hot paths
- DocumentFragment optimization: Batch DOM updates for plan rendering
- Timer cleanup: Clear notification/countdown timers on SSE reconnect
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add plan-orchestrator.ts with parallel subagent spawning
- Phase 1: 4 specialist subagents analyze in parallel (requirements, architecture, testing, risks)
- Phase 2: Synthesis merges and deduplicates outputs
- Phase 3: Verification subagent assigns priorities and identifies gaps
- New /api/generate-plan-detailed endpoint with SSE progress updates
- Auto-regenerate plan when switching between Standard/Enhanced modes
- Show quality score, warnings, and gaps from verification
- Rename "Detailed" to "Enhanced" button with tooltip
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add /api/sessions/:id/ralph-prompt/write endpoint to write prompt to @ralph_prompt.md
- Wizard now writes full prompt to file, then sends simple read command to Claude
- Fix session readiness check to wait for prompt character instead of notWorking flag
- Add E2E test infrastructure for ralph-loop workflow (port 3190)
- Add ralph-wizard-prod.mjs script for production testing
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix screen creation failing when using cpulimit/nice (env vars were passed
as command args instead of shell variables)
- Fix session status never becoming 'idle' after startInteractive()
- Add automatic state file cleanup on startup (removes orphaned sessions)
- Add /api/cleanup-state endpoint for manual cleanup
- Make frontend session readiness check more robust (20 attempts, multiple checks)
- Add fallback to direct PTY write if writeViaScreen fails
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
CPU Limiting:
- Add CpuLimitConfig type and settings in App Settings
- Use nice command for process priority (-20 to 19)
- Optional cpulimit integration for hard CPU limits (if installed)
- Applied to new sessions via screen-manager and session modules
Ralph Loop Wizard:
- 3-step wizard: Describe → Plan → Launch
- New /api/generate-plan endpoint generates implementation steps via Claude
- Plan editor with checkboxes, priorities (P0/P1/P2), reordering
- Skip plan option for manual workflows
- Compact wizard modal layout
UI Updates:
- Form section headers for grouped settings
- CPU limit status indicator (cpulimit availability)
- Wizard modal wider layout and inline fields
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
API Endpoints:
- GET /api/sessions/:id/fix-plan - Generate fix plan markdown from todos
- POST /api/sessions/:id/fix-plan/import - Import todos from markdown
- POST /api/sessions/:id/fix-plan/write - Write @fix_plan.md to working dir
- POST /api/sessions/:id/fix-plan/read - Read and import from working dir
RalphTracker Methods:
- generateFixPlanMarkdown() - Creates structured markdown with priority sections
- importFixPlanMarkdown() - Parses markdown and imports todos with priorities
UI Features:
- Menu dropdown in Ralph panel (hamburger icon)
- View Fix Plan modal with copy/write buttons
- Import from file button
- Reset Circuit Breaker button
- Fix plan modal with syntax-highlighted textarea
Bump version to 0.1392
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Smart Respawn Integration (2.5):
- Check circuit breaker OPEN state before respawn (blocks cycle)
- Check RALPH_STATUS EXIT_SIGNAL=true to stop respawn
- Check STATUS=BLOCKED to pause respawn for human intervention
- Use RECOMMENDATION from RALPH_STATUS in update prompt
- Emit respawn:blocked SSE event with reason and details
- Show notification when respawn is blocked
- Visual blocked state with pulsing red indicator
Priority-based Todos (part of 2.3):
- Add RalphTodoPriority type (P0/P1/P2/null)
- Parse priority from todo content (P0:, Critical:, High Priority, etc.)
- Add priority field to RalphTodoItem interface
- Update frontend to display priority badges with colors
- Sort todos by priority (P0 > P1 > P2) then by status
- Add CSS for priority badges and highlighted cards
Bump version to 0.1391
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Phase 1 Ralph enhancement implementation:
- Parse RALPH_STATUS blocks from Claude output (status, tasks, files, tests)
- Circuit breaker state machine (CLOSED → HALF_OPEN → OPEN) for stuck detection
- Dual-condition exit gate (completion indicators >= 2 AND EXIT_SIGNAL)
- Frontend: circuit breaker badge (yellow warning, red stuck), status block display
- API: circuit breaker reset endpoint, ralph status endpoint
- Notifications when circuit breaker opens or exit gate is met
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add getLightState() and getLightSessionsState() methods
- SSE init payload reduced from ~1.9MB to ~5KB (~400x smaller)
- Terminal buffers are fetched on-demand when switching tabs
- Fixes UI becoming unresponsive on SSE reconnect
chore: bump version to 0.1388
- Session creation now parallelized (3x faster for multiple tabs)
- Minimized agents badge: compact, hover-triggered, auto-hide
- Removed "Minimized Agents" header, status dots instead of text
- Click to pin dropdown open, hover for quick access
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Session Manager:
- Store event handlers for proper cleanup on session stop
- Add resetSessionManager() for test isolation
State Store:
- Use atomic write pattern (temp file + rename) to prevent corruption
- Add tokenStats to AppState interface
Subagent Watcher:
- Clean up pendingToolCalls on agent completion
- Add cleanupStaleAgents() for 24h+ old agents
- Clear all state maps on stop() for clean restart
Web Server:
- Remove listeners from transcript watcher before stopping
- Clean up respawn timers when session is deleted
- Use cleanupSession() in spawn orchestrator for proper resource cleanup
Frontend:
- Close existing EventSource before reconnecting to prevent duplicates
- Close failed connection before scheduling reconnect
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add max 500k tokens per session limit (Claude's context is ~200k)
- Add max 100k tokens per update to prevent sudden jumps from parsing errors
- Reject "M" suffix values > 0.5M to prevent false matches in output text
- Add validation to restoreTokens() to reject corrupted values from state.json
- Add validation to addToGlobalStats() for negative and absurd values
- Fix daily sessions count: track unique session IDs instead of incrementing
on every 5-minute recording interval
- Pass sessionId to recordDailyUsage() for accurate session counting
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Instead of timing-based correlation between terminal output and subagent
files, now reads the Task tool's description parameter directly from the
parent session's transcript. This provides 100% reliable mapping between
agent IDs and their short descriptions, fixing issues where multiple
agents spawned in quick succession would get the same window title.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add parseTaskDescriptionsFromTerminalData() to parse raw terminal data in
interactive sessions (not just one-shot mode)
- Strip ANSI codes before regex matching since terminal output has embedded
codes like [1mExplore[0m
- Fix getTaskDescriptionForSubagent() to check ALL sessions with matching
project hash (not just the first one found) since multiple sessions can
share the same working directory
- Add tool_result event forwarding for subagent SSE broadcasts
The pattern Explore(Description) in terminal output is now correctly parsed
and used as the subagent window title instead of extracting from the prompt.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Extract short description from TaskTracker (params.description) for subagent windows
- Correlate subagents with sessions by working directory and timing
- Auto-minimize subagent windows when agents complete
- Persist minimized/open window states to server for cross-browser sync
- Restore window states on reconnection/page reload
- New API: GET/PUT /api/subagent-window-states
- State file: ~/.claudeman/subagent-window-states.json
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Subagents (Task tool background agents) are autonomous and cannot
receive input mid-execution. They complete and return results to
the parent session. Removed the input UI and API endpoint.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When sessions were restored after server restart, lastRecordedTokens
was not initialized, causing the periodic token recording to treat
the entire restored token count as new usage. This inflated daily
stats by re-counting tokens on every server restart.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Prevents false idle detection when server is rebuilt/restarted.
Previously, restored respawn controllers would start immediately
and trigger after 30s of "no output" because the session wasn't
fully connected yet.
Now waits 2 minutes from server startup before activating restored
respawn controllers, giving sessions time to reconnect and stabilize.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add 300-char rolling window to catch working patterns split across PTY chunks
- Check completion message BEFORE working patterns (priority fix)
- Clear rolling window on completion message (transition point)
- Increase working pattern absence threshold from 3s to 8s
- Add Session.isWorking safety check before confirming idle
- Add 20+ more working patterns (Compiling, Building, Processing, etc.)
- Make AI idle checker prompt more conservative (err toward WORKING)
- Update documentation
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The run summary tracker wasn't tracking respawn events for restored
sessions because the tracker didn't exist when setupRespawnListeners
was called during session restoration. Changed to lazy lookup so
the tracker is retrieved each time an event fires.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements a "what happened while you were away" view for sessions:
- Track significant events: session start/stop, respawn cycles, state changes,
idle/working transitions, token milestones, auto-compact/clear, Ralph
completions, AI check results, hook events, errors/warnings
- Stats dashboard: respawn cycles, peak tokens, active/idle time, issue counts
- Timeline view with color-coded severity and emoji icons
- Filter events by type (all/errors/warnings/respawn/idle)
- State stuck detection (warns when state unchanged for 10+ minutes)
Click the chart icon on any session tab to open the Run Summary modal.
Files:
- src/run-summary.ts: RunSummaryTracker class with event tracking
- src/types.ts: RunSummary, RunSummaryEvent types
- src/web/server.ts: Integration with session/respawn events, API endpoint
- src/web/public/: Modal UI, timeline rendering, filter controls
- docs/run-summary-plan.md: Implementation plan
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implement multi-phase improvement to respawn controller's idle detection:
Phase 1 - Stop Hook Detection:
- Add signalStopHook() method - definitive signal when Claude finishes
- 3s confirmation timer to handle race conditions
- Skip AI check when hook received (100% confidence)
Phase 2 - idle_prompt Detection:
- Add signalIdlePrompt() method - fires after 60s+ of idle
- Immediately confirms idle (no confirmation timer needed)
Phase 3 - Transcript File Monitoring:
- New TranscriptWatcher class watches session JSONL files
- Detects: completion, tool execution, plan mode, errors
- Signals respawn controller for supporting detection
Web UI Updates:
- Hook indicator with purple styling and pulse animation
- Shows "Stop hook received" or "idle_prompt hook received"
- 100% confidence displayed with hook-confirmed style
This significantly improves idle detection reliability by using
definitive signals from Claude Code rather than parsing terminal output.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add TokenUsageEntry and TokenStats types for daily tracking
- Add state-store methods for recording and retrieving daily usage
- Add /api/token-stats endpoint with daily breakdown and totals
- Add token stats modal with summary cards, 7-day bar chart, and table
- Record tokens periodically (5min) and on session delete
- Make header tokens clickable to open stats modal
- Fix subagents panel toggle icon direction (down when open)
- Increase subagents panel height when expanded
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add 100ms debounce for new file detection to allow content to be written
- Add subagent:updated event for retroactive description updates
- Extract description in processEntry when first user message is processed
- Add extractDescriptionFromFile helper with retry in file change handler
- Update SubagentTranscriptEntry.content type to support string | array
- Add test coverage for subagent:updated event
Fixes 43% failure rate where subagents displayed raw IDs instead of descriptions
due to race condition when files were discovered before first line was written.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Move COM shorthand to prominent position after safety warning
- Add version sync note (must match package.json)
- Consolidate Key Files into 4 logical categories
- Clarify next available test port (3128)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Extract better titles from task description (first line/sentence)
- Make subagent window text white for better readability
- Expand monitor panel to 700px when subagents tab is active
- Add smooth width transition when switching tabs
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add SubagentWatcher service that discovers and tails Claude Code's subagent
transcript files (~/.claude/projects/.../subagents/agent-*.jsonl)
- Track tool calls, progress events, and messages in real-time via file watching
- Add SSE events for subagent:discovered, subagent:tool_call, subagent:progress,
subagent:message, subagent:completed
- Add REST API endpoints: GET /api/subagents, /api/subagents/:id,
/api/subagents/:id/transcript
- Add subagent panel in Monitor section showing active agents and their activity
- Add floating draggable windows that auto-open for each active agent
- Add standalone CLI tool: scripts/watch-subagents.ts for terminal monitoring
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add AiCheckerBase abstract class to eliminate ~700 lines of duplication
between AiIdleChecker and AiPlanChecker
- Fix missing persistence fields (completionConfirmMs, noOutputTimeoutMs)
that caused custom timing values to be lost on server restart
- Extract restoreRespawnController() helper to dedupe session restoration
logic between state.json and screens.json paths
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When /respawn/stop was called, the controller was stopped but remained
in the respawnControllers map. This caused persistSessionState() to
still read config.enabled=true and save respawnEnabled=true to state.json.
On server restart, the respawn controller would be re-enabled because
state.json had respawnEnabled=true.
Fix: Remove controller from map and clear any timed respawn when
/respawn/stop is called, so persistSessionState() correctly sets
respawnEnabled=false.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- persistSessionState() now saves respawnEnabled based on config.enabled
(user intent) instead of controller.state (running state)
- cleanupSession() saves respawn config BEFORE removing controller
- AiIdleChecker and AiPlanChecker constructors filter undefined values
to prevent overwriting defaults (fixes "timed out after undefinedms")
- updateConfig() methods also filter undefined values
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add restoreTokens() method to Session class for recovery after restart
- Add GlobalStats type for cumulative usage tracking across all sessions
- Accumulate tokens from deleted sessions into global stats
- Track lifetime session count with incrementSessionsCreated()
- Add /api/stats endpoint for global stats
- Include globalStats in /api/status response
- Frontend shows aggregate tokens + cost in header
- Fix respawn controller config to filter undefined values
- Add 7 new tests for global stats functionality (v0.1343)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Display version number in center of bottom toolbar
- Load version from package.json via server's /api/status
- Version shown on initial load and SSE reconnect
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix "enable twice" bug: add enabled:true to getModalRespawnConfig()
- Add timer tracking with countdown display in respawn banner
- New events: timerStarted, timerCancelled, timerCompleted, actionLog
- Show active timers with progress bars and recent action log
- Skip remaining timeouts after AI check returns IDLE/WORKING verdict
- Enhanced DetectionStatus with activeTimers, recentActions, currentPhase
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds a two-stage gate before auto-accepting plan mode prompts:
1. Strict regex pre-filter - checks for numbered options + selector
2. AI confirmation - spawns Opus to classify as PLAN_MODE or NOT_PLAN_MODE
This prevents spurious Enter key presses when Claude is paused mid-thought
or experiencing network lag, rather than showing a plan approval prompt.
New files:
- src/ai-plan-checker.ts: AI checker following ai-idle-checker.ts pattern
Config fields added to RespawnConfig:
- aiPlanCheckEnabled (default: true)
- aiPlanCheckModel (default: claude-opus-4-5-20251101)
- aiPlanCheckMaxContext (default: 8000)
- aiPlanCheckTimeoutMs (default: 60000)
- aiPlanCheckCooldownMs (default: 30000)
New events: planCheckStarted, planCheckCompleted, planCheckFailed
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
arkon