FN-2910 surfaced concurrent reviewer + merger activity on the same task.
Root cause: asymmetric in-flight guards let an unpause-resume kick off a
fresh executor session while a recovery path was already running, and the
auto-merge handoff fired before the executor's finally block finished
cleanup. This sweeps the surrounding lifecycle paths for similar races and
tightens the reviewer pause gate against TOCTOU through runtime setup.
- Symmetric in-flight tracking across `executing`, `recoveringCompleted`,
and `resumingUnpaused`; `recoverCompletedTask` bails when any are set.
- Atomic claim of the recovery slot in the completed-task watchdog before
any awaited work.
- Workflow-rerun bounce returns "bounced" | "skipped-pending" so the
watchdog can no longer log a false-success retry when the original
bounce is still mid-flight.
- Self-healing's completed-task scan re-checks executing IDs inside the
loop instead of trusting a pre-await snapshot.
- 300ms grace period before auto-merge enqueue, giving the executor's
finally block (session disposal, child cleanup) time to drain and
eliminating the residual log-overlap symptom from FN-2910. Test uses
fake timers, no real sleep added.
- New AgentSemaphore.runNested for synchronously nested helper agents
(reviewers): bumps activeCount for honest observability while bypassing
the wait queue, preserving forward-progress fairness for the parent at
low maxConcurrent. Both createReviewStepTool and triage's
createReviewSpecTool now use it.
- New beforeSpawnSession hook on AgentRuntimeOptions/AgentOptions fired
inside createFnAgent immediately before createAgentSession, past every
awaited setup step. Reviewer wires a pause re-check that throws a
sentinel error converted to UNAVAILABLE, closing the TOCTOU window
where pause flipped during runtime resolution or resource loading.
All 2887 engine tests pass; engine + core + cli + dashboard + plugin-sdk
+ pi-claude-cli + desktop typecheck clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer subprocesses were spawned via fn_review_spec / fn_review_step
even with globalPause on, because reviewer.ts had no pause awareness.
Stuck detector also kept running, treating pause-disposed sessions as
inactivity and re-queuing tasks. Pause-transition listeners only called
session.dispose(), which doesn't always interrupt an in-flight LLM
stream — letting reviewer spawns leak through after pause flipped.
- reviewer.ts: re-read settings, return UNAVAILABLE without spawning
when globalPause/enginePaused is on.
- stuck-task-detector.ts: skip checkStuckTasks() while paused.
- triage.ts / executor.ts: call session.abort() before dispose() in the
pause-transition listener to interrupt in-flight work.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Fix triage planning model resolution to fall back through project/global planning settings and default overrides
- Fix reviewer model selection to honor validator-specific settings before default provider/model overrides
- Update merger model resolution to apply default override fallback and align shared task setting types/executor flow
- Add regression coverage for triage, reviewer, and merger fallback behavior and update settings hierarchy documentation
Triage with claude-sonnet-4-6 via pi-claude-cli kept looping on
fn_review_spec calls that were rejected by pi's validator with
"root: must be object". Parameterless MCP tools (schema
{type:"object", properties:{}}) emit zero input_json_delta events,
so partialJson stayed "" and the catch fell through to
finalArgs = "" — a string, which TypeBox's Type.Object({}) rightly
refuses. Default empty partialJson to {} so the call lands.
Also:
- Add a 2-step reminder loop in triage before swapping to the
fallback planning model — primary models that wrote PROMPT.md
but forgot fn_review_spec recover from a nudge, no need to pay
the cold-start tax of a new triage on a different model.
- Inject @runfusion/fusion's own pi extension into dashboard/
daemon/serve sessions and propagate the path to createFnAgent
via setHostExtensionPaths so fn_* tools register globally
without requiring `pi install npm:@runfusion/fusion`.
- Drop the "historical" qualifier from replayed tool labels —
Claude was reading "TOOL RESULT (historical Read):" as
"previous session, ignore" and looping on verification.
- Remove subprocess-lifecycle stderr debug logs that landed for
hang diagnosis — root cause is fixed, the noise can go.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add formatError() helper that extracts both message and full stack from
unknown caught values, and use it at every status:"failed" catch site in
executor, agent-heartbeat, and triage. Stack traces now land in
store.logEntry outcome (persisted to task.log/activityLog) and in stderr
logger output, so failures like "Cannot read properties of undefined
(reading 'filter')" can be diagnosed without re-running.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Sort eligible todo tasks in Scheduler by priority first, then createdAt/id for stable FIFO ordering within each tier
- Sort eligible triage tasks using the same priority-aware ordering while preserving existing pause/status/recovery gating
- Add regression tests for scheduler and triage ordering, including blocked/paused/recovery-gated edge cases
- Update architecture docs to describe priority-first task dispatch behavior
- Update Fusion skill docs, prompts, and capability references to use public fn_* tool names consistently
- Align engine system prompts and tool schemas for messaging/task actions with fn_send_message, fn_read_messages, fn_task_* naming
- Refresh related tests across CLI, engine, dashboard, and core to match normalized tool naming and behavior
- Add a patch changeset for @runfusion/fusion describing the skill-tool namespace normalization
Atomic worktree creation: on `git worktree add` failure, remove the
partial directory so retries see a clean slate (executor.ts,
step-session-executor.ts). Add `reapOrphanWorktrees()` sweep on engine
startup to delete `.worktrees/*` dirs that have no `.git` file and are
not registered with git. Force readonly workflow steps to run pre-merge
so review personas reuse the coding worktree instead of allocating a
fresh post-merge one. Inject a Frontend UX Criteria checklist into
TRIAGE_SYSTEM_PROMPT when a task touches dashboard UI, so the first
coding pass can meet UX reviewer expectations.
Motivated by FN-2185 post-mortem: an incomplete `.worktrees/pale-raven`
dir blocked retries, and three Step 6 revision cycles over cosmetic UX
issues consumed ~13 hours.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Root cause: during a triage split the AI could set a child task's
`dependencies` to the parent id. The parent is hard-deleted after the split,
and the scheduler's dep check treats a missing id as unmet — permanently
blocking the dependent. This stranded FN-2164 behind the deleted FN-2163.
- core/store.deleteTask: refuse to delete when any live task still has the id
in its `dependencies` array. Throws TaskHasDependentsError listing dependents
so callers can rewrite or recover. Covers the triage-split path and any
future caller.
- engine/triage task_create: validate each proposed dependency before creating
a child — reject the parent id, reject unknown task ids, allow siblings
created earlier in the same split or pre-existing tasks.
- engine/triage split cleanup: wrap the parent deleteTask in try/catch that
keeps the parent alive (safer than stranding dependents) and logs the reason.
- engine/triage prompts: both the mandatory-split and proactive-split prompts
now explicitly state that subtask deps must never reference the parent.
- dashboard/routes /subtasks/create-tasks: reject parent-id deps, drop unknown
deps with an audit log entry, surface parentTaskCloseError + droppedDependencies
in the response instead of silently swallowing them.
- engine/executor: on execute entry, detect the drift state (in-progress task
with no worktree) and emit a loud log + task log entry; the existing
fresh-worktree path then recovers. Prevents silent "operating without a
worktree" behavior that we saw on FN-2152.
Tests:
core: 2907/2907 pass (+5 new, incl. deleteTask guard regression)
engine: 2554/2554 pass (+17 new, incl. task_create dep validation)
dashboard: 9064/9064 pass (+2 new for /subtasks/create-tasks).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Rename engine export and call sites to use createFnAgent consistently across runtime flows
- Update core lazy engine loader and dashboard agent-generation/planning/chat paths to reference createFnAgent
- Refresh affected unit and integration tests, including renaming pi-create-kb-agent.test.ts to pi-create-fn-agent.test.ts
- Update AGENTS.md documentation references to match the new createFnAgent name
- Add MailboxView as a full-page navigation view replacing the modal-based approach
- Migrate MessageStore from filesystem to SQLite backend for message persistence
- Implement conversation grouping for inbox display with unread badge state
- Remove modal plumbing (useModalManager, AppModals exports) and simplify App.tsx
- Add evictStaleProcessing() to TriageProcessor for self-healing hung triage sessions
- Add comprehensive MailboxView tests and Header mailbox tests
- Update README documentation with MailboxView features
- Add CSS styles for MailboxView component
- Fix MobileNavBar tests for mailbox tab visibility
- executor.test.ts: remove unused imports (Column, StuckTaskDetector),
replace Function type with EventListener, add MockTaskStore interface
- restart.integration.test.ts: replace require() with ESM import,
replace Function types with proper function signatures
- All tests pass
- Remove unused imports across 25 files in engine package
- Remove unused variable declarations in ipc-worker.ts, child-process-runtime.ts, and mission-autopilot.ts
- Clean up unnecessary imports in agent-instructions.ts, agent-tools.ts, cron-runner.ts, executor.ts, and other modules
- Minor cleanup in notifier.ts, peer-exchange-service.ts, pi.ts, plugin-runner.ts, and other files
- Improves code quality and reduces potential confusion from unused code
- Add SkillSelectionResolver and SkillSelectionContext for deterministic skill sets
- Implement skill-resolver.ts with request/filter/resolve workflow
- Add session-skill-context.ts helper for unified skill selection across all session types
- Wire skill override into createKbAgent via AgentOptions
- Apply skill selection to triage, executor, reviewer, merger, and heartbeat sessions
- Add comprehensive tests for skill-resolver and session-skill-context
- Export skill resolver types and functions from engine index
- Add skill selection resolver pattern to memory documentation
- Restructure PluginManager detail view with card-based layout
- Add mobile responsive styles with adaptive grid
- Add desktop CSS with card-based grid design
- Clear triage status when task moves to todo
- Add changeset for plugin settings design
After a process crash or restart, tasks retain "specifying" status in the
database but no agent is actually running. These stale statuses consume
concurrency slots, blocking all new triage work. Added a startup sweep
that clears orphaned "specifying" statuses, mirroring the existing
merge status cleanup.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The triage processor was only checking the global semaphore, not the
per-project maxConcurrent setting. Now it counts all active agents
(in-progress + specifying) and respects Math.min(perProjectAvailable,
semaphoreAvailable) before starting new triage tasks. This matches the
scheduler's concurrency enforcement and prevents triage from consuming
all available slots on startup.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The triage processor was kicking off all eligible tasks simultaneously,
queuing them all on the semaphore regardless of available slots. Now it
checks semaphore.availableCount first and only starts as many triage
tasks as there are free slots. Remaining tasks get picked up on the
next poll cycle (default 10-15s) as slots free up.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add ESLint configuration (eslint.config.mjs) for TypeScript/JavaScript linting
- Update executor prompts to include lint instruction before code submission
- Add lint check to triage prompt validation workflow
- Update agent prompts to emphasize lint compliance as quality requirement
- Add lint tool to agent toolset with file-level rule disabling capability
- Include lint in CI workflow with non-blocking status
- Update tests to verify lint-inclusive prompt behavior
- Add documentation for lint integration in contributing.md
- Add changeset for @gsxdsm/fusion minor release
- Update executor, triage, and merger prompt strings to identify the product as "fn"
- Align branch/worktree documentation examples with fusion/fn-* naming conventions
- Rename dashboard websocket attachment guard from __kbWebSocketsAttached to __fnWebSocketsAttached
- Refresh engine tests, dashboard utility comments, and TUI header text to remove stale kb wording
- Add task_document_write and task_document_read factories with schemas, revision-aware responses, and error handling in agent-tools
- Wire document tools into executor and triage sessions and update prompts to persist and reuse planning artifacts across runs
- Export document tool factories and parameter schemas from engine public entrypoints while keeping executor re-export compatibility
- Add comprehensive tests covering write/read success paths, empty/not-found cases, and store failure handling
- 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 planningModelProvider/planningModelId fields to task model and API batch-update endpoint
- Wire triage engine to use per-task planning model override with settings hierarchy fallback
- Add planning model selector to TaskDetailModal Model tab
- Add comprehensive tests for triage engine planning model resolution and dashboard API/routes
- Add AgentPromptTemplate and AgentPromptsConfig types to ProjectSettings
- Create agent-prompts module with 7 built-in prompt templates and role resolver
- Wire engine agents (executor, reviewer, merger, triage) to use resolved prompts
- Add 25 test cases covering template resolution, role assignment, and validation
- Export new types from @fusion/core package
- Document agentPrompts configuration and built-in templates in AGENTS.md
- Add project-memory module in @fusion/core with read/write/resolve helpers and upsert-on-store hook
- Bootstrap project memory during task creation (store) with structured task context
- Inject resolved project memory into executor and triage agent system prompts
- Add comprehensive tests for project-memory module, store integration, and agent prompt changes
- Document project memory architecture and usage in README
- Update triage agent to proactively suggest splitting large tasks into subtasks during specification
- Tighten reviewer guidance to flag undersplit tasks that should be broken down further
- Add comprehensive tests for proactive subtask creation in triage (192 lines)
- Add reviewer tests for undersplit task detection (12 lines)
- Update README with documentation on proactive subtask splitting behavior
- Include user comments as context during AI triage spec generation and spec review
- Invalidate stale spec approvals when new user comments are added after approval
- Add lastApprovedAt tracking to task metadata for approval freshness detection
- Add comprehensive tests for comment-aware triage, review, and stale approval logic
- Add changeset for published package and update README with feature documentation
