macro-agent 0.0.11 → 0.0.13

package/.macro-agent/teams/self-driving/prompts/grinder.md

@@ -0,0 +1,27 @@
+# Grinder
+
+You are a Grinder — an autonomous worker that claims tasks and executes them.
+
+## How You Work
+
+1. Call claim_task() to get your next task
+2. Read the task description carefully
+3. Execute the work: write code, run tests, fix issues
+4. When done, call done() with your results
+5. Immediately call claim_task() for the next task
+6. If no tasks available, wait briefly and retry
+
+## Work Guidelines
+
+- Work independently — do not ask for instructions
+- Write complete implementations, not stubs or TODOs
+- Run tests before calling done() if the task involves code changes
+- Commit your changes with clear, descriptive messages
+- If you're stuck on a task for more than 10 minutes, call done({ status: "blocked" })
+
+## Quality
+
+- Follow existing codebase conventions
+- Ensure your changes compile and pass lint
+- Write tests for new functionality
+- Keep changes focused on the task — don't refactor surrounding code

package/.macro-agent/teams/self-driving/prompts/judge.md

@@ -0,0 +1,27 @@
+# Judge
+
+You periodically evaluate the health of the codebase and take corrective action.
+
+## Evaluation Cycle
+
+Every time you activate:
+1. Run the build — check compilation
+2. Run the test suite — check correctness
+3. Run the linter — check code quality
+4. If all pass: emit GREEN_SNAPSHOT
+5. If any fail: create fixup tasks with tag "fixup" and priority "critical"
+
+## Creating Fixup Tasks
+
+When you find failures:
+- Create one task per distinct issue (don't bundle)
+- Include the exact error output in the task description
+- Tag with: ["fixup", "<subsystem>"]
+- Reference the failing file and test
+
+## Health Reporting
+
+After each evaluation cycle, emit a health status:
+- Use emit_status to report overall codebase health
+- Include: build status, test pass rate, lint error count
+- This information helps the planner adjust priorities

package/.macro-agent/teams/self-driving/prompts/planner.md

@@ -0,0 +1,33 @@
+# Planner
+
+You are the Planner. Your job is to continuously explore the codebase,
+understand the current state, and create well-defined tasks for workers.
+
+## How You Work
+
+1. **Explore**: Read the codebase to understand architecture, patterns, and gaps
+2. **Plan**: Break the objective into independent, parallelizable tasks
+3. **Create**: Use create_task to add tasks to the pool with clear descriptions and tags
+4. **Monitor**: Watch for completed/failed tasks and adjust the plan
+5. **Repeat**: Planning is continuous — as work completes, create the next batch
+
+## Planning Guidelines
+
+- Each task should be completable in 5-30 minutes by a single worker
+- Tag tasks with subsystem and type for filtered claiming
+- Set dependencies (blockers) when ordering matters
+- Prefer many small tasks over few large ones — parallelism is the goal
+- Include enough context for a worker to start without re-exploring
+
+## Constraints
+
+- Do NOT instruct on things the model already knows (coding, testing, etc.)
+- DO specify things specific to this codebase (conventions, build system, deploy pipeline)
+- Constraints are more effective than instructions: "No TODOs, no partial implementations"
+
+## Sub-Planners
+
+For large subsystems (>10 tasks), spawn a sub-planner:
+- spawn_agent({ role: "planner", task: "Plan the [subsystem] changes" })
+- Sub-planners create tasks in the shared pool
+- You maintain the high-level view

package/.macro-agent/teams/self-driving/roles/grinder.yaml

@@ -0,0 +1,17 @@
+name: grinder
+extends: worker
+display_name: "Grinder"
+description: "Claims and executes tasks autonomously"
+
+capabilities_add:
+  - task.claim
+  - git.push
+
+prompt: prompts/grinder.md
+
+macro_agent:
+  lifecycle:
+    type: ephemeral
+    task_bound: false
+    max_duration_ms: 3600000
+    self_cleanup: true

package/.macro-agent/teams/self-driving/roles/judge.yaml

@@ -0,0 +1,24 @@
+name: judge
+extends: monitor
+display_name: "Judge"
+description: "Periodically evaluates codebase health and creates fixup tasks"
+
+capabilities_add:
+  - exec.build
+  - exec.test
+  - exec.lint
+  - task.create
+  - task.update
+  - git.branch.create
+  - git.push
+
+prompt: prompts/judge.md
+
+macro_agent:
+  workspace:
+    type: own
+    branch_pattern: "judge/{agent-id}"
+    cleanup_on_terminate: true
+  lifecycle:
+    type: event-driven
+    parent_bound: false

package/.macro-agent/teams/self-driving/roles/planner.yaml

@@ -0,0 +1,18 @@
+name: planner
+extends: coordinator
+display_name: "Planner"
+description: "Continuously explores the codebase, creates and prioritizes tasks"
+
+capabilities_add:
+  - task.claim
+
+capabilities_remove:
+  - agent.spawn.integrator
+  - agent.spawn.monitor
+
+prompt: prompts/planner.md
+
+macro_agent:
+  lifecycle:
+    type: daemon
+    cascade_terminate: true

package/.macro-agent/teams/self-driving/team.yaml

@@ -0,0 +1,103 @@
+name: self-driving
+description: "Autonomous codebase development with continuous planning"
+version: 1
+
+roles:
+  - planner
+  - grinder
+  - judge
+
+topology:
+  root:
+    role: planner
+    prompt: prompts/planner.md
+    config:
+      model: sonnet
+
+  companions:
+    - role: judge
+      prompt: prompts/judge.md
+      config:
+        model: haiku
+
+  spawn_rules:
+    planner: [grinder, planner]
+    judge: []
+    grinder: []
+
+communication:
+  channels:
+    task_updates:
+      description: "Task lifecycle events"
+      signals: [TASK_CREATED, TASK_COMPLETED, TASK_FAILED]
+    work_coordination:
+      description: "Work assignment and completion"
+      signals: [WORK_ASSIGNED, WORKER_DONE, MERGE_REQUEST]
+    health:
+      description: "System health monitoring"
+      signals: [HEALTH_CHECK, METRIC_SNAPSHOT, GREEN_SNAPSHOT]
+
+  subscriptions:
+    planner:
+      - channel: task_updates
+      - channel: work_coordination
+        signals: [WORKER_DONE]
+      - channel: health
+        signals: [METRIC_SNAPSHOT]
+    judge:
+      - channel: task_updates
+        signals: [TASK_FAILED]
+      - channel: work_coordination
+        signals: [WORKER_DONE]
+      - channel: health
+    grinder:
+      - channel: work_coordination
+        signals: [WORK_ASSIGNED]
+
+  emissions:
+    planner: [TASK_CREATED, WORK_ASSIGNED]
+    judge: [HEALTH_CHECK, GREEN_SNAPSHOT, FIXUP_CREATED]
+    grinder: [WORKER_DONE]
+
+  routing:
+    status: upstream
+    peers:
+      - from: judge
+        to: planner
+        via: direct
+        signals: [FIXUP_CREATED, GREEN_SNAPSHOT]
+      - from: planner
+        to: judge
+        via: direct
+        signals: [CONVERGENCE_CHECK]
+
+  enforcement: permissive
+
+macro_agent:
+  task_assignment:
+    mode: pull
+    pull:
+      idle_timeout_s: 300
+      claim_retry_delay_ms: 2000
+      max_concurrent_per_agent: 1
+
+  integration:
+    strategy: trunk
+    config:
+      max_retries: 3
+      conflict_action: abandon
+
+  lifecycle:
+    continuations:
+      enabled: true
+      max_history_messages: 50
+      checkpoint_interval: round_trip
+    scaling:
+      min_workers: 3
+      max_workers: 20
+      scale_on: task_queue_depth
+      idle_drain: true
+
+  observability:
+    metrics_window_s: 3600
+    snapshot_interval_s: 300

package/.macro-agent/teams/structured/prompts/developer.md

@@ -0,0 +1,26 @@
+# Developer
+
+You are a **Developer** on a structured development team. You execute a single assigned task in your isolated workspace.
+
+## Responsibilities
+
+1. **Implement** - Complete the assigned task according to its description
+2. **Test** - Verify your changes work correctly
+3. **Commit** - Create clean, descriptive commits
+4. **Report** - Call `done()` with status and summary when finished
+
+## Workflow
+
+1. Read and understand the task description
+2. Explore the relevant code to plan your changes
+3. Implement the changes
+4. Run tests to verify correctness
+5. Commit your work with a clear message
+6. Call `done()` with status `completed` and a brief summary
+
+## Guidelines
+
+- Stay focused on your assigned task only
+- Do not modify files outside the scope of your task
+- If you encounter a blocker, call `done()` with status `failed` and describe the issue
+- Make atomic commits that are easy to review

package/.macro-agent/teams/structured/prompts/lead.md

@@ -0,0 +1,25 @@
+# Lead
+
+You are the **Lead** of a structured development team. You coordinate work by decomposing objectives into tasks and assigning them to developers.
+
+## Responsibilities
+
+1. **Task Decomposition** - Break down the user's request into well-scoped, independent tasks
+2. **Assignment** - Spawn developers and assign each a specific task
+3. **Progress Tracking** - Monitor task completion and handle failures
+4. **Sequencing** - Ensure dependent tasks are ordered correctly using blockers
+
+## Workflow
+
+1. Analyze the objective and identify the minimal set of changes needed
+2. Create tasks with clear descriptions and acceptance criteria
+3. Spawn developers to handle each task
+4. Monitor TASK_COMPLETED / TASK_FAILED signals
+5. When all tasks complete, call `done()` with a summary
+
+## Guidelines
+
+- Keep tasks small and focused (one file or one logical change)
+- Set blockers between tasks that have data dependencies
+- Prefer parallel execution where tasks are independent
+- If a task fails, assess whether to retry or adjust the plan

package/.macro-agent/teams/structured/prompts/reviewer.md

@@ -0,0 +1,24 @@
+# Reviewer
+
+You are a **Reviewer** on a structured development team. You review merge requests to ensure quality before they land on the main branch.
+
+## Responsibilities
+
+1. **Review** - Examine code changes in merge requests for correctness and quality
+2. **Validate** - Run build and test to verify the changes don't break anything
+3. **Decide** - Approve or reject the merge request with clear reasoning
+
+## Workflow
+
+1. Wait for MERGE_REQUEST signals
+2. Check out the branch and review the diff
+3. Run `build` and `test` commands
+4. If everything passes, emit REVIEW_APPROVED
+5. If issues found, emit REVIEW_REJECTED with explanation
+
+## Guidelines
+
+- Focus on correctness, not style preferences
+- Run the test suite before approving
+- Provide specific, actionable feedback on rejections
+- Do not modify code directly; report issues back to the lead

package/.macro-agent/teams/structured/roles/developer.yaml

@@ -0,0 +1,12 @@
+name: developer
+extends: worker
+display_name: "Developer"
+description: "Executes assigned tasks in isolated workspaces"
+
+prompt: prompts/developer.md
+
+macro_agent:
+  lifecycle:
+    type: ephemeral
+    task_bound: true
+    self_cleanup: true

package/.macro-agent/teams/structured/roles/lead.yaml

@@ -0,0 +1,11 @@
+name: lead
+extends: coordinator
+display_name: "Lead"
+description: "Decomposes work into tasks and assigns them to developers"
+
+prompt: prompts/lead.md
+
+macro_agent:
+  lifecycle:
+    type: persistent
+    cascade_terminate: true

package/.macro-agent/teams/structured/roles/reviewer.yaml

@@ -0,0 +1,19 @@
+name: reviewer
+extends: monitor
+display_name: "Reviewer"
+description: "Reviews merge requests and approves or rejects changes"
+
+capabilities_add:
+  - exec.build
+  - exec.test
+
+prompt: prompts/reviewer.md
+
+macro_agent:
+  workspace:
+    type: own
+    branch_pattern: "review/{agent-id}"
+    cleanup_on_terminate: true
+  lifecycle:
+    type: event-driven
+    parent_bound: false

package/.macro-agent/teams/structured/team.yaml

@@ -0,0 +1,89 @@
+name: structured
+description: "Hierarchical development with explicit task assignment and merge queue"
+version: 1
+
+roles:
+  - lead
+  - developer
+  - reviewer
+
+topology:
+  root:
+    role: lead
+    prompt: prompts/lead.md
+    config:
+      model: sonnet
+
+  companions:
+    - role: reviewer
+      prompt: prompts/reviewer.md
+      config:
+        model: haiku
+
+  spawn_rules:
+    lead: [developer, lead]
+    reviewer: []
+    developer: []
+
+communication:
+  channels:
+    task_updates:
+      description: "Task lifecycle events"
+      signals: [TASK_CREATED, TASK_ASSIGNED, TASK_COMPLETED, TASK_FAILED]
+    merge_flow:
+      description: "Merge queue and integration events"
+      signals: [MERGE_REQUEST, MERGE_COMPLETED, MERGE_CONFLICT]
+    review_flow:
+      description: "Code review signals"
+      signals: [REVIEW_REQUESTED, REVIEW_APPROVED, REVIEW_REJECTED]
+
+  subscriptions:
+    lead:
+      - channel: task_updates
+      - channel: merge_flow
+      - channel: review_flow
+        signals: [REVIEW_APPROVED, REVIEW_REJECTED]
+    reviewer:
+      - channel: merge_flow
+        signals: [MERGE_REQUEST]
+      - channel: review_flow
+    developer:
+      - channel: task_updates
+        signals: [TASK_ASSIGNED]
+
+  emissions:
+    lead: [TASK_CREATED, TASK_ASSIGNED]
+    reviewer: [REVIEW_APPROVED, REVIEW_REJECTED]
+    developer: [MERGE_REQUEST, TASK_COMPLETED, TASK_FAILED]
+
+  routing:
+    status: upstream
+    peers:
+      - from: reviewer
+        to: lead
+        via: direct
+        signals: [REVIEW_APPROVED, REVIEW_REJECTED]
+
+  enforcement: permissive
+
+macro_agent:
+  task_assignment:
+    mode: push
+
+  integration:
+    strategy: queue
+    config:
+      require_review: true
+
+  lifecycle:
+    continuations:
+      enabled: false
+    scaling:
+      min_workers: 1
+      max_workers: 10
+      scale_on: manual
+      idle_drain: false
+
+  observability:
+    metrics_window_s: 1800
+    snapshot_interval_s: 600

package/.sudocode/issues.jsonl

@@ -147,3 +147,9 @@
 {"id":"i-3ojo","uuid":"1677a703-8b6a-40ef-a86c-2e311730d924","title":"Phase 3: Conversation Lifecycle — Create on spawn, close on done/terminate","content":"\nAutomatically create conversations on agent spawn / user session init, close on done() / terminate, track agent-to-conversation mapping.\n\n## Changes\n\n### Create\n- `src/mail/conversation-map.ts` — Agent-to-conversation mapping (agentConversationMap + peerConvIndex)\n\n### Modify\n- `src/agent/agent-manager.ts` — In spawn(): create task conversation, join parent+child, map agent. In terminate(): close conversation, remove from map.\n- `src/mcp/tools/done.ts` — Between step 4 (emitStatus) and step 5 (dispatchDone): record completion turn, close conversation\n- `src/api/server.ts` — In POST /api/init: create session conversation. In POST /api/conversation/message: record user/assistant turns.\n- `src/server/combined-server.ts` — Wire ConversationMap to services\n\n## Key Details\n- ConversationMap is in-memory, reconstructible from EventStore on restart\n- Peer index keyed by sorted agent pair [min(A,B), max(A,B)]\n- All mail operations wrapped in try/catch — never fail core operations\n- done() insertion point: between step 4 (line ~234) and step 5 (line ~236)\n- Head manager participates in session conversation AND each child's task conversation\n\n## Verification\n- Unit test: spawn → conversation created with participants. done() → turn recorded + closed. terminate → closed.\n- Unit test: API init → session conversation. User message → turns recorded.\n- Existing agent-manager, done(), API server tests pass\n","status":"closed","priority":1,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-06 05:10:09","updated_at":"2026-02-06 05:43:01","closed_at":"2026-02-06 05:43:01","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-3ojo","from_type":"issue","to":"i-44q8","to_type":"issue","type":"blocks"},{"from":"i-3ojo","from_type":"issue","to":"s-gu8h","to_type":"spec","type":"implements"}],"tags":["lifecycle","mail","phase-3"],"feedback":[{"id":"57de1bdb-243a-48f4-9e89-7d1406318b83","from_id":"i-3ojo","to_id":"s-gu8h","feedback_type":"comment","content":"Phase 3 implemented successfully. Key deviations from plan:\n- Added `setMailServices()` late-binding method to AgentManager interface (AgentManager is created before combined-server creates MailService)\n- Fixed EventStore `applyConversationEvent` to map arbitrary `close_reason` to valid `ConversationStatus` (previously used close_reason directly as status)\n- Both standalone (`createAPIServer`) and shared (`createAPIApp`) API modes updated with session conversation creation and turn recording\n- 15 new tests covering ConversationMap, session/task conversations, done/terminate cleanup, and full lifecycle flow","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-06T05:43:09.098Z","updated_at":"2026-02-06T05:43:09.098Z"}]}
 {"id":"i-44q8","uuid":"e99900fc-6f62-49e7-9f8b-a1f3c150c510","title":"Phase 4: Turn Recording — MessageRouter hook with hierarchy-aware resolution","content":"\nAdd turnRecorder callback to MessageRouter so direct messages automatically become conversation turns, with hierarchy-aware conversation resolution and peer conversation auto-creation.\n\n## Changes\n\n### Create\n- `src/mail/turn-recorder.ts` — Conversation resolution + turn recording callback\n\n### Modify\n- `src/router/message-router.ts` — Add turnRecorder to config, setTurnRecorder() for late-binding, invoke after event emission + wake handler for agent and task addresses only\n- `src/router/types.ts` — Add TurnRecorder type\n- `src/server/combined-server.ts` — Wire turnRecorder to MessageRouter\n\n## Resolution Algorithm\n1. Only agent and task addresses produce turns (return null for scope/role/broadcast/siblings)\n2. Get both agents' lineage from EventStore\n3. Parent→child or child→parent: use child's task conversation\n4. Peers: getOrCreatePeerConversation() — sorted-pair key, both joined, parent = nearest common ancestor\n5. Record turn with sourceType: 'intercepted', sourceMessageId\n\n## Key Details\n- Follow wakeHandler callback pattern in MessageRouterConfig\n- setTurnRecorder() enables late-binding (services created after router)\n- Try/catch wrapping — never fail message delivery due to turn recording\n- Hook points: agent address after line 333, task address after line 403\n\n## Verification\n- Unit test each resolution case: parent→child, child→parent, peer, broadcast (no turn)\n- Integration test: full hierarchy, send messages, verify conversation tree\n- Existing MessageRouter tests pass\n","status":"closed","priority":1,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-06 05:10:19","updated_at":"2026-02-06 05:46:25","closed_at":"2026-02-06 05:46:25","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-44q8","from_type":"issue","to":"i-7614","to_type":"issue","type":"blocks"},{"from":"i-44q8","from_type":"issue","to":"s-gu8h","to_type":"spec","type":"implements"}],"tags":["mail","phase-4","router"],"feedback":[{"id":"8f9d5fa9-ba8c-4355-99e7-7690b648c860","from_id":"i-44q8","to_id":"s-gu8h","feedback_type":"comment","content":"Phase 4 implemented. Added TurnRecorderCallback type to router/types.ts, setTurnRecorder() late-binding method on MessageRouter, turn recording hooks for agent and task addresses only (not scope/role/broadcast/hierarchical), createTurnRecorder() with hierarchy-aware resolution (parent↔child → child's task conversation, peers → auto-created peer conversation). 9 new tests covering parent→child, child→parent, peer auto-creation, peer reuse, task address, and edge cases.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-06T05:46:25.563Z","updated_at":"2026-02-06T05:46:25.563Z"}]}
 {"id":"i-7614","uuid":"44846f3e-829c-41bf-a4bc-b7e73b3de759","title":"Phase 5: External Access — REST endpoints and WebSocket channels","content":"\nREST API endpoints for querying conversations/turns, WebSocket channels for real-time conversation updates, mail-backed conversation history.\n\n## Changes\n\n### Modify\n- `src/api/types.ts` — Add ConversationSummary, TurnSummary, WSTurnAdded, WSConversationUpdate types\n- `src/api/server.ts` — Add REST endpoints (GET /api/conversations, GET /api/conversations/:id, GET /api/conversations/:id/turns, POST /api/conversations/:id/close). Add WebSocket channels (conversation:${id}, conversations). Update GET /api/conversation/history to use mail-backed data when enabled.\n\n## Key Details\n- Follow existing /api/agents, /api/tasks endpoint patterns\n- WebSocket channels driven by EventStore onConversationChange() and onTurnChange() listeners\n- GET /api/conversation/history falls back to flat conversationHistory[] when mail disabled\n\n## Verification\n- Unit test each REST endpoint with mocked services\n- Integration test: full flow through REST API\n- WebSocket test: subscribe → send message → verify turn_added notification\n- Existing API tests pass\n","status":"closed","priority":2,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-06 05:10:25","updated_at":"2026-02-06 05:51:39","closed_at":"2026-02-06 05:51:39","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-7614","from_type":"issue","to":"s-gu8h","to_type":"spec","type":"implements"}],"tags":["api","mail","phase-5"],"feedback":[{"id":"eca59751-c0e3-4b27-a84c-46a25b216060","from_id":"i-7614","to_id":"s-gu8h","feedback_type":"comment","content":"Phase 5 implemented. Added REST endpoints: GET /api/conversations (list with type/status filter + pagination), GET /api/conversations/:id (detail), GET /api/conversations/:id/turns (paginated), POST /api/conversations/:id/close, GET /api/conversations/:id/participants. Added WebSocket channels: 'conversations' (global conversation updates), 'conversation:${id}' (per-conversation updates + turn_added events). Added conversation_update and turn_added to WSMessageType. Used shared registerConversationRoutes() to avoid duplication between standalone and shared server modes. 14 new tests for the REST endpoints.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-06T05:51:38.798Z","updated_at":"2026-02-06T05:51:38.798Z"}]}
+{"id":"i-96f6","uuid":"3f23e84b-1a2e-416e-87a7-d28c3a6e0a98","title":"Wire peer routing from communication config","content":"\n`setupPeerSubscriptions()` in `team-runtime.ts:418-427` ignores `manifest.communication.routing.peers` config entirely. It hardcodes mutual subtree subscriptions only between root and companion agents.\n\n**What needs to happen:**\n- `TeamRuntime.bootstrap()` reads `manifest.communication.routing.peers`\n- Resolves role names to spawned agent IDs\n- Sets up subscriptions based on `via` field:\n  - `via: \"direct\"` → mutual agent-to-agent subscriptions\n  - `via: \"topic\"` → ensure both agents subscribe to named topic\n  - `via: \"scope\"` → scope-based addressing\n- Per-peer `signals` filter stored for later filtering\n\n**Impact:** Without this, non-hierarchical communication patterns defined in team YAML have no runtime effect. Only root↔companion hardcoded wiring works.\n\n**Files:** `src/teams/team-runtime.ts`\n","status":"closed","priority":2,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-07 21:26:55","updated_at":"2026-02-09 03:22:26","closed_at":"2026-02-09 03:22:26","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-96f6","from_type":"issue","to":"s-29pg","to_type":"spec","type":"implements"}],"tags":["communication","gap","teams"],"feedback":[{"id":"4a4af7e4-8354-41cf-8eb2-d70457b41c70","from_id":"i-96f6","to_id":"s-29pg","feedback_type":"comment","content":"Implemented config-driven peer routing in `src/teams/team-runtime.ts`:\n\n**Core changes:**\n- `wirePeerRoutes()` reads `manifest.communication.routing.peers` and wires subscriptions based on `via` field\n- `wireSinglePeerRoute()` maps `via: \"direct\"` → subtree sub, `via: \"topic\"` → shared topic sub, `via: \"scope\"` → role channel sub\n- Each `PeerConnection` creates a **directional** (one-way) subscription\n- Falls back to legacy bidirectional subtree subs when no `routing.peers` config exists\n\n**Deferred wiring:**\n- Roles not spawned at bootstrap (e.g., grinder) are tracked in `pendingPeerRoutes`\n- `setupDeferredPeerWiring()` uses `onLifecycleEvent(\"spawned\")` to wire pending routes when roles become available\n- Auto-unsubscribes when all pending routes are resolved\n\n**Signal filter storage:**\n- In-memory `peerSignalFilters` map (key: `fromAgentId→toAgentId`, value: signal names)\n- `getPeerSignalFilters()` getter exposes for future signal filtering (issue `i-3o8g`)\n- Peer routes serialized into `team_config` EventStore event as `peerRoutes` field for cross-process access\n\n**Tests:** 8 new tests in `team-system.test.ts` covering config-driven routing, legacy fallback, signal filter storage, deferred wiring, teardown cleanup, topic via, and scope via.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-09T03:22:22.796Z","updated_at":"2026-02-09T03:22:22.796Z"}]}
+{"id":"i-4dh7","uuid":"2192c5c8-2a0c-4846-919a-843f638ce633","title":"Add wake logic to routeStatusToTopicSubscribers()","content":"\n`routeStatusToTopicSubscribers()` in `message-router.ts` delivers status notifications to topic co-subscribers but does **not** call `wakeHandler`. Sleeping agents won't be proactively woken by lifecycle events from peers.\n\n**What needs to happen:**\n- Add `wakeHandler` calls in `routeStatusToTopicSubscribers()` for each notified agent\n- Mirror the pattern used in `sendToAddress()` scope handling (lines 506-523)\n- Wrap in try/catch per existing wake pattern\n\n**Note:** `routeStatusToSubtreeSubscribers()` also lacks wake logic — this is consistent. Consider adding wake to both or just topic routing.\n\n**Files:** `src/router/message-router.ts`\n","status":"closed","priority":2,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-07 21:26:59","updated_at":"2026-02-09 04:26:24","closed_at":"2026-02-09 04:26:24","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-4dh7","from_type":"issue","to":"s-29pg","to_type":"spec","type":"implements"}],"tags":["communication","gap","router","teams"],"feedback":[{"id":"52f15b2a-8421-4ffe-a7b7-c4502796b106","from_id":"i-4dh7","to_id":"s-29pg","feedback_type":"comment","content":"Added wake logic to both status routing functions in `src/router/message-router.ts`:\n\n- `routeStatusToSubtreeSubscribers()` — now calls `wakeHandler` after emitting each status notification to subtree subscribers\n- `routeStatusToTopicSubscribers()` — same for topic co-subscribers\n\n**Design choices:**\n- Priority: `\"normal\"` — wakes idle/sleeping agents but queues for busy (prompting) ones. `\"low\"` was considered but it never wakes by design, defeating the purpose.\n- Wake errors are caught with try/catch — never fail status delivery due to wake handler issues\n- Follows the exact pattern from `sendToAddress()` scope handling\n\n**Tests:** 4 new tests in `message-router.test.ts`: wake on subtree status, wake on topic status, no wake when busy, resilience to wake handler errors.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-09T04:26:24.094Z","updated_at":"2026-02-09T04:26:24.094Z"}]}
+{"id":"i-3o8g","uuid":"e9f34e0e-a6b2-41cf-bfed-e6ab0cf7187e","title":"Implement signal filtering for topic subscriptions","content":"\n`ChannelSubscription.signals?: string[]` allows per-role filtering of which signals to receive on a channel. Currently `getTopicsForRole()` in `team-runtime.ts:329-341` extracts channel names but discards the signals array entirely. All agents on a topic receive all signals.\n\n**What needs to happen:**\n- Store per-agent signal filters alongside topic subscriptions (in EventStore or in-memory lookup)\n- `routeStatusToTopicSubscribers()` checks the filter before delivering — skip delivery when the status signal doesn't match the agent's configured filter\n- Option 1 (recommended for Phase 1): Filter in `check_messages` MCP tool\n- Option 2: Filter in router during delivery\n\n**Files:** `src/teams/team-runtime.ts`, `src/router/message-router.ts`\n","status":"closed","priority":3,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-07 21:27:04","updated_at":"2026-02-09 05:02:02","closed_at":"2026-02-09 05:02:02","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-3o8g","from_type":"issue","to":"i-96f6","to_type":"issue","type":"depends-on"},{"from":"i-3o8g","from_type":"issue","to":"s-29pg","to_type":"spec","type":"implements"}],"tags":["communication","gap","teams"],"feedback":[{"id":"2439b9de-6335-4016-acd0-ac1afb796dce","from_id":"i-3o8g","to_id":"s-29pg","feedback_type":"comment","content":"Implemented signal filtering for both subtree and topic status routing.\n\n**Approach**: Late-bound `SignalFilter` callback on MessageRouter (follows `setTurnRecorder` pattern). TeamRuntime installs a combined filter that checks two sources:\n\n1. **Peer connection filters** (per-agent-pair, directional): From `routing.peers[].signals`. Checked first — takes precedence.\n2. **Channel subscription filters** (per-role): From `subscriptions[role][].signals`. Pre-computed into a `roleAllowedSignals` map at install time. If any subscription for a role has no `signals` array, the role receives all signals.\n\n**Key decisions**:\n- Signal name carried in `details.signal` field of status events (not `status_type`)\n- Untagged events (no `details.signal`) always pass through for backwards compatibility\n- Filter installed once at bootstrap; deferred wiring updates `agentRoleMap` but doesn't reinstall the filter (the closure captures the maps by reference)\n\n**Files modified**: `src/router/message-router.ts` (SignalFilter type, setSignalFilter, filter checks in both routing functions), `src/teams/team-runtime.ts` (installSignalFilter, agentRoleMap)\n\n**Tests**: 5 new router tests (suppress, allow, passthrough, subtree, directional args) + 7 new team tests (install, peer allow/block, untagged passthrough, channel sub filter, all-signals role, peer precedence). All 3,081 tests pass.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-09T05:01:57.319Z","updated_at":"2026-02-09T05:01:57.319Z"}]}
+{"id":"i-96z9","uuid":"13867978-72d1-4e65-b99d-5a0f0e762f37","title":"Add test coverage for monitorContinuations()","content":"\n`monitorContinuations()` in `team-runtime.ts` has zero test coverage. The method watches agent lifecycle events and auto-resumes daemon agents on unexpected stop.\n\n**Test cases needed:**\n1. Daemon agent stops unexpectedly → `continueAgent()` called after 1s delay\n2. Agent stops with \"completed\" status → no continuation triggered\n3. Agent stops with \"cancelled\" status → no continuation triggered\n4. Continuation spawn failure → error caught, no crash\n5. Multiple lifecycle events for same agent → no duplicate continuations\n6. Teardown unsubscribes → no continuation after teardown\n\n**Edge cases:**\n- State consistency when agent ID changes after continuation\n- Rapid stop/restart cycles\n\n**Files:** `src/teams/__tests__/team-system.test.ts` or new `continuation.test.ts`\n","status":"closed","priority":2,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-07 21:27:09","updated_at":"2026-02-08 08:08:45","closed_at":"2026-02-08 08:08:45","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-96z9","from_type":"issue","to":"s-5w36","to_type":"spec","type":"implements"}],"tags":["continuations","teams","testing"],"feedback":[{"id":"ff29220b-52ce-4b49-b720-5971ce0fb376","from_id":"i-96z9","to_id":"s-5w36","feedback_type":"comment","content":"Implemented 6 test cases for `monitorContinuations()` in `src/teams/__tests__/team-system.test.ts`:\n1. Auto-continue root agent on unexpected stop\n2. Auto-continue companion agent on unexpected stop\n3. No continuation on completed status stop\n4. No continuation on cancelled status stop\n5. No trigger for non-monitored agents\n6. Unsubscribe on teardown prevents continuation\n\nAll tests passing.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-08T08:08:36.918Z","updated_at":"2026-02-08T08:08:36.918Z"}]}
+{"id":"i-3y08","uuid":"1c832346-74f6-4683-a01d-1dc3af035277","title":"Wire TrunkStrategy conflictAction and strategy lifecycle hooks","content":"\nTwo minor gaps in the integration strategy subsystem:\n\n1. **`conflictAction` unused:** `TrunkIntegrationStrategy` stores `conflictAction` config (\"abandon\" | \"queued_for_resolution\") but `land()` always returns `status: \"conflict\"` without differentiating. Should set `action` field on the conflict result based on config.\n\n2. **`initialize()`/`close()` lifecycle hooks never called:** The `IntegrationStrategy` interface defines optional `initialize()` and `close()` methods but neither TeamRuntime nor the worker handler invokes them. TeamRuntime should call `initialize()` during `initialize()` and `close()` during `teardown()`.\n\n**Files:** `src/workspace/strategies/trunk.ts`, `src/teams/team-runtime.ts`\n","status":"closed","priority":3,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-07 21:27:15","updated_at":"2026-02-08 08:08:46","closed_at":"2026-02-08 08:08:46","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-3y08","from_type":"issue","to":"s-6ok3","to_type":"spec","type":"implements"}],"tags":["gap","integration-strategies","teams"],"feedback":[{"id":"fb19956e-2987-4024-8bbc-23a8282af3f3","from_id":"i-3y08","to_id":"s-6ok3","feedback_type":"comment","content":"Both gaps addressed:\n\n1. **conflictAction wired**: Added `action` field to `LandResult` type in `src/workspace/strategies/types.ts`. `TrunkIntegrationStrategy.land()` now sets `action` on both conflict return paths, mapping config's `\"abandon\"` to result's `\"abandoned\"`.\n\n2. **Strategy lifecycle hooks called**: `TeamRuntime.initialize()` now instantiates the strategy via `defaultStrategyRegistry.get()` and calls `strategy.initialize()`. `TeamRuntime.teardown()` calls `strategy.close()`. Both wrapped in try/catch for best-effort semantics. Added `getIntegrationStrategy()` getter.\n\nFiles modified: `src/workspace/strategies/types.ts`, `src/workspace/strategies/trunk.ts`, `src/teams/team-runtime.ts`","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-08T08:08:41.672Z","updated_at":"2026-02-08T08:08:41.672Z"}]}
+{"id":"i-1zso","uuid":"3573f69b-89c2-499f-ba0c-faaa578b5165","title":"Implement emission restrictions and enforcement mode","content":"\nTwo related features that are loaded/stored but not enforced at runtime:\n\n1. **Emission restrictions:** `communication.emissions` maps roles to allowed signals. Loader validates role names, but no code checks whether an agent's emitted signal is in its allowed list.\n\n2. **Enforcement mode:** `communication.enforcement` (\"strict\" | \"permissive\" | \"audit\") stored in EventStore `team_config` event. Never retrieved or branched on.\n\n**Implementation per A10 in implementation-details.md:**\n- Enforcement at MCP tool level (not router): `emit_status` tool handler checks team manifest emissions for agent's role\n- `strict`: reject disallowed emissions\n- `permissive`: log warning, allow\n- `audit`: record in EventStore for analysis\n\n**Files:** `src/mcp/mcp-server.ts` or `src/mcp/tools/`, `src/router/message-router.ts`\n","status":"closed","priority":3,"assignee":null,"archived":0,"archived_at":null,"created_at":"2026-02-07 21:27:20","updated_at":"2026-02-09 05:45:05","closed_at":"2026-02-09 05:45:05","parent_id":null,"parent_uuid":null,"relationships":[{"from":"i-1zso","from_type":"issue","to":"i-3o8g","to_type":"issue","type":"depends-on"},{"from":"i-1zso","from_type":"issue","to":"s-29pg","to_type":"spec","type":"implements"}],"tags":["communication","enforcement","gap","teams"],"feedback":[{"id":"9abcdf0a-c065-4814-8a0a-db59d15b8ae2","from_id":"i-1zso","to_id":"s-29pg","feedback_type":"comment","content":"Implemented emission restrictions and enforcement mode.\n\n**Approach**: Late-bound `EmissionValidator` callback on MessageRouter (follows `setSignalFilter` pattern). TeamRuntime installs the validator during bootstrap.\n\n**How it works**:\n1. `emitStatus()` extracts `details.signal` from the status event\n2. Calls the validator with `(agentId, signal)` — validator looks up agent's role via `agentRoleMap`, checks against `communication.emissions[role]`\n3. Returns action based on enforcement mode:\n   - `strict` → `reject` (blocks emission entirely, `emitStatus` returns early)\n   - `permissive` → `warn` (allows through)\n   - `audit` → `audit` (records audit event in EventStore with `emission_violation` type, then allows through)\n4. Untagged events (no `details.signal`) and unknown agents always pass through\n\n**Data serialization**: Added `emissions` to the team_config EventStore event for cross-process access by MCP subprocesses.\n\n**Files modified**: `src/router/message-router.ts` (EmissionValidator type, setEmissionValidator, validation in emitStatus), `src/teams/team-runtime.ts` (installEmissionValidator, emissions in team_config)\n\n**Tests**: 5 new router tests (reject, allow, audit with event recording, warn, untagged passthrough) + 8 new team tests (install, allow, strict reject, permissive warn, audit, untagged, unknown agent, no-config skip, serialization). All 3,095 tests pass.","agent":"alexngai","anchor":null,"dismissed":false,"created_at":"2026-02-09T05:45:00.325Z","updated_at":"2026-02-09T05:45:00.325Z"}]}