- Add skills discovery API (GET /api/skills/discovered) to list available skills with enabled state
- Add skills execution toggle API (PATCH /api/skills/execution) for enabling/disabling skills with project-scoped persistence
- Add skills catalog API (GET /api/skills/catalog) with resilient fallback to fetch skills.sh catalog
- Skills are stored in project settings (.fusion/settings.json) with support for both top-level and package-scoped skills
- Add SkillsAdapter runtime class for skills discovery, catalog fetching, and execution toggle
- Add comprehensive tests for all skills API endpoints
- Update dashboard, serve, and provider-settings commands with skills adapter integration
- Skip flaky streamChatResponse test (matches main branch behavior)
An AI review agent (FN-1506) killed the running dashboard by finding
the process on port 4040 via lsof and running kill -9, causing exit
code 137 (SIGKILL) with no logs. This adds multi-layer guardrails:
- AGENTS.md: project-level rule reserving port 4040
- Executor/reviewer system prompts: explicit prohibition on killing
port 4040 processes, with instruction to use --port 0 instead
- Core agent-prompts.ts: same guardrails in all prompt variants
- Reviewer told to issue REVISE if executor violates the rule
- SIGHUP handlers in dashboard.ts and serve.ts for resilience
- Background engine reconciliation in dashboard/serve startup
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove the anti-pattern where the cwd project was treated as "primary" with a
special engine, and other projects got "secondary" engines through a separate
code path. Every project now gets an identical ProjectEngine created through
ProjectEngineManager.
Key changes:
- Add ProjectEngineManager class to @fusion/engine for uniform engine lifecycle
- Replace manual engine maps in dashboard.ts and serve.ts with engineManager
- Add engineManager to ServerOptions for per-project engine resolution
- Add getProjectContext() helper in routes.ts (replaces 199 getScopedStore calls)
- Merge and automation routes now resolve engine subsystems per-request
- SSE endpoint uses engine's store when available (same EventEmitter)
- Fix tsx not found in dev-with-memory.mjs startup script
- Add invalidateAllGlobalSettingsCaches for cross-project settings sync
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Secondary projects were started via ProjectManager (bare InProcessRuntime)
which lacks auto-merge queue, startup sweep, periodic retry, PR monitor,
and settings listeners. Tasks reaching in-review in secondary projects
would never be auto-merged. Now all projects use ProjectEngine for the
full subsystem set.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Tasks in projects other than the primary (cwd) project were never triaged
because only one ProjectEngine was started. When a project is accessed via
?projectId= API/SSE, getOrCreateProjectStore created a TaskStore but left
the Scheduler, TriageProcessor, and TaskExecutor unstarted.
Fix: introduce setOnProjectFirstCreated callback in project-store-resolver
so the dashboard server is notified when any new project is first accessed.
dashboard.ts creates a ProjectManager that lazily starts an InProcessRuntime
(Scheduler + TriageProcessor + TaskExecutor) for each project the first time
it is accessed — works for any number of registered projects.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add process lifecycle diagnostics for dashboard and serve commands
- Add SQLite database health check to diagnostics endpoint
- Add store listener count diagnostics for debugging subscription leaks
- Audit and fix SSE connection management to prevent connection leaks
- Audit and fix timer/interval cleanup in engine and CLI shutdown handlers
- Fix res.on() call guard for test mocks compatibility
- Fix variable declaration ordering in serve.ts
- Update memory with diagnostic findings for future debugging
Implemented:
- POST /api/missions/features/:featureId/validate - triggers validation run
- GET /api/missions/features/:featureId/validation-loop - returns loop snapshot
- GET /api/missions/features/:featureId/validation-runs - returns run history with pagination
- GET /api/missions/validation-runs/:runId - returns run detail with assertion results
- POST /api/missions/recover - triggers recovery of active missions
- SSE events for milestone:validation:updated via assertion CRUD and link/unlink
All 4xx/5xx responses use consistent {"error": "message"} format.
Uses existing badRequest, notFound, internalError helpers.
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
- Implement actual AI response parsing in parseValidationResult() with JSON extraction
from markdown code blocks, repair for common JSON issues, and assertion result parsing
- Fix notifyValidationComplete to pass feature.taskId instead of featureId to
handleTaskCompletion() in in-process-runtime, dashboard, and serve
- Fix recoverActiveMissions() to actually transition validating features back to
implementing and call processTaskOutcome for features with completed tasks
- Add comprehensive unit tests for MissionExecutionLoop lifecycle, processTaskOutcome,
recoverActiveMissions, and error handling
Broaden the auto-heal pattern to cover build-verification failures and
add a 30m idle cooldown so tasks that exhausted mergeRetries without
matching the narrow heal pattern get another sweep-driven attempt
instead of being stranded until a human clears the counter.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Deterministic verification failures were embedding the raw stderr/stdout
(up to 50MB per VERIFICATION_COMMAND_MAX_BUFFER) in a second log entry,
flooding logs/stdout and crashing the app. The runVerificationCommand
helper already wrote a truncated summary, so verifyDeterministicBuild
now just references it.
When the failure surfaces in the dashboard merge handler, kick the task
back to in-progress with a steering comment so the agent can fix the
failing test/build instead of parking it in in-review with a fatal
error.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Bundles staged work-in-progress modifications across multiple packages
(routes, store, agent-instructions, self-healing, QuickEntryBox, etc.)
plus the dashboard theme-data.css preload fix.
Note: an unstaged 621-line deletion in .fusion/memory.md was deliberately
NOT committed — it appears to be an accidental overwrite of architecture
notes and is left in the working tree for review.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Wrap AppInner with NodeProvider in App.tsx
- Read node context state (currentNode, currentNodeId, isRemote, setters)
- Add useRemoteNodeData and useRemoteNodeEvents hooks
- Sync selected node with useNodes() results
- Update Board/ListView/ExecutorStatusBar to use remote tasks when in remote mode
- Add mocks for NodeContext, useRemoteNodeData, useRemoteNodeEvents, useNodes in App.test.tsx
- Clear fusion-dashboard-current-node in test setup to avoid cross-test leakage
- Add explicit boundary comments to serve.ts and dashboard.ts clarifying semaphore lane usage
- Add regression tests for semaphore lane-vs-utility boundary in serve.test.ts and dashboard.test.ts
- Add changeset for @gsxdsm/fusion patch release
- Add MemoryInsights class in @fusion/core for AI-powered memory audit generation
- Add post-run hook to CronRunner for triggering memory summarization after scheduled tasks
- Wire memory background processing in both dashboard and serve commands
- Add memoryAuditEnabled and memoryAuditSchedule settings for configurable automation
- Fix startup ordering: sync automation before cronRunner.start() to prevent race conditions
- Add comprehensive tests for memory-insights and dashboard/serve integration
- Update contributing.md and settings-reference.md with documentation
The cachedMaxConcurrent variable driving the AgentSemaphore was only
updated during unpause transitions and merge retry timers. Changing
maxConcurrent in the dashboard had no effect on the semaphore until one
of those events fired, causing fewer tasks to run than configured.
Also adds HeartbeatMonitor and HeartbeatTriggerScheduler wiring to both
serve and dashboard commands, with tests.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Require gh-authenticated paths for task pr-create and dashboard PR routes
- Switch dashboard command and extension GitHub tools to use the gh auth client instead of token-based auth
- Update CLI and dashboard tests to cover gh-only createPr and import behavior
- Refresh README guidance and add a changeset removing the GitHub token requirement for PR flows
- Wrap dashboard auth storage with API-key provider helpers derived from model registry providers
- Normalize provider display names and bridge set/clear/has API-key operations to AuthStorage credentials
- Expand route and onboarding tests for mixed OAuth/API-key states and API-key-authenticated setup paths
- Stabilize assignment-trigger heartbeat timing test by replacing fixed delays with waitFor assertions
- Wrap TaskExecutor task:updated async listener logic in a top-level guard and log uncaught listener failures
- Add explicit catch handling for ProjectManager activity logging and dashboard stuck-detector settings-triggered checkNow calls
- Document async EventEmitter guard conventions in scheduler and hybrid executor listener wiring
- Add regression tests for executor and dashboard listener guards and include a patch changeset for @gsxdsm/fusion
- Update CLI command strings, help text, and project-resolution messaging to reference fn
- Sync core backup/store messaging and dashboard API/planning/subtask prompt examples with fn usage
- Refresh dashboard UI copy in settings, project detection, and PR guidance for consistent command naming
- Adjust CLI/dashboard tests and add a @gsxdsm/fusion changeset for the naming alignment patch
- Fix pause-abort handler using updateTask({status: undefined}) which was a
no-op, leaving tasks stuck in 'specifying' forever after a pause interrupts
a session post-APPROVE. Changed to status: null to actually clear the field.
- Apply same fix to transient-error retry and general error catch paths.
- Wire StuckTaskDetector into TriageProcessor: trackTask/untrackTask/recordActivity
on session lifecycle, markStuckAborted to prevent stuck kills from being
reported as errors, and clear status to null on stuck-kill for next-poll retry.
- Update dashboard.ts to pass stuckTaskDetector to TriageProcessor and call
triageRef.current?.markStuckAborted in the shared onStuck callback.
- Add 3 tests covering pause-abort status clearing and stuck detector wiring.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add instructionsPath and instructionsText fields to Agent type and AgentStore
- Create agent-instructions resolver module in engine with priority-based resolution
- Wire custom instructions into executor, triage, reviewer, and merger agents
- Add PATCH /agents/:id/instructions API endpoint with file and text support
- Add instructions editor UI to dashboard agent detail config tab
- Add comprehensive tests for instructions resolver and AgentStore integration
- Add changeset for published package bump
- Add companies.sh type definitions and parser in @fusion/core with comprehensive tests
- Create CLI `kb agent-import` command to import agents from companies.sh
- Add POST /api/agents/import dashboard endpoint with dry-run support
- Build AgentImportModal component with search, preview, and batch import UI
- Add dashboard route tests for the agent import API endpoint
- Create changeset for @gsxdsm/fusion minor bump
- Define AiPromptExecutor type and update CronRunner constructor to accept injected executor
- Implement executeAiPromptStep with real agent session execution in worktrees
- Create createAiPromptStepExecutor factory and wire up in dashboard CLI
- Add comprehensive tests for AI prompt step execution (mock agent, errors, output capture)
- Fix mock typing for dashboard test build compatibility
- Instantiate MissionAutopilot with TaskStore and MissionStore before Scheduler creation
- Pass missionAutopilot and missionStore to Scheduler constructor options
- Break circular dependency via setScheduler() after Scheduler construction
- Start autopilot in engine startup sequence, stop on graceful shutdown
- Add comprehensive tests verifying MissionAutopilot wiring in both test files
- Wire AgentStore into TaskExecutor initialization within dashboard.ts startup
- Add AgentStore mock to dashboard command tests for proper test coverage
- Update test mocks to reflect new AgentStore dependency injection
- Add ContextLimitDetector to detect agent loops via repeated tool call patterns
- Implement compact-and-resume strategy: summarize conversation and restart agent from current step
- Add loop recovery to StuckTaskDetector with configurable attempt tracking and retry limits
- Extend executor with automatic loop recovery on context limit detection
- Add loop recovery support to pi executor with same compact-and-resume pattern
- Add comprehensive tests for context-limit-detector, stuck-task-detector loop detection, executor, and pi recovery
- Add changeset for patch bump to @gsxdsm/fusion
- Update README with loop detection and recovery documentation
When the stuck task detector killed a task and immediately called
moveTask("todo"), the scheduler could re-dispatch the task before the
old execution's finally block cleared this.executing. The new execute()
call hit the guard and silently returned, stranding the task in
"in-progress" with no active session or worktree (seen on FN-810/FN-912).
Move the requeue responsibility from StuckTaskDetector.killAndRetry to
the executor's finally block, which runs after this.executing.delete().
The beforeRequeue budget check now runs before session.dispose() and its
result is passed via StuckTaskEvent.shouldRequeue → markStuckAborted().
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add explicit no-progress loop detection signals to executor agent sessions
- Enhance stuck-task detector with configurable strategies (timeout, no-progress, combined)
- Add comprehensive test coverage for stuck-task detector (371 lines of tests)
- Remove modelFilter test utilities (198 lines of dead test code)
- Simplify modelFilter.ts by removing unused filtering logic
- Clean up dashboard server startup and minor README fix
- Add openrouterModelSync setting to eagerly fetch OpenRouter model catalog at
dashboard startup so the model picker shows all available models
- Add toggle in Settings → Models to disable the sync
- Add mission-interview session management module for AI-guided mission specs
- Include tests for the sync opt-out behavior
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Thread real project context (projectName, projectId) into ntfy notification deep links
- Dashboard deep-link URL now honors project context with /project/:name/task/:id pattern
- Update App.tsx routing to handle project-scoped deep link navigation
- Add project info to NtfyNotifier types and wire through notification payload
- Add changeset for published @gsxdsm/fusion package
- Add tests for notifier project context and dashboard routing
fix: update scheduler filesystem path from .kb to .fusion for task validation
style: clean up TaskDetailModal styling with reusable CSS classes
test: add comprehensive tests for file-service operations with mocked filesystem
- Add project API methods and types (fetchProjects, registerProject, fetchProjectHealth, etc.)
- Add ProjectCard component with health metrics, status badges, and pause/resume actions
- Add server-side project management routes for multi-project orchestration
- Add ActivityFeed component with grouped entries and project badges
- Add SetupWizard component with 5-step project creation flow
- Fix TypeScript errors in server routes for listProjects and getGlobalConcurrencyState
