Codewalkers

tas/Codewalkers

Fork 0

Commit Graph

Author	SHA1	Message	Date
Lukas May	fab7706f5c	feat: Phase schema refactor, agent lifecycle module, and log chunks Phase model changes: - Drop `number` column (ordering now by createdAt + dependency DAG) - Replace `description` (plain text) with `content` (Tiptap JSON) - Add `approved` status as dispatch gate - Add phase dependency management (list, remove, dependents) - Approval gate in PhaseDispatchManager.queuePhase() Agent log chunks: - New `agent_log_chunks` table for DB-first output persistence - LogChunkRepository port + DrizzleLogChunkRepository adapter - FileTailer onRawContent callback streams chunks to DB - getAgentOutput reads from DB first, falls back to file Agent lifecycle module (src/agent/lifecycle/): - SignalManager: atomic signal.json read/write/wait operations - RetryPolicy: exponential backoff with error-specific strategies - ErrorAnalyzer: pattern-based error classification - CleanupStrategy: debug archival vs production cleanup - AgentLifecycleController: orchestrates retry/recovery flow - Missing signal recovery with instruction injection Completion detection fixes: - Read signal.json file instead of parsing stdout as JSON - Cancellable pollForCompletion with { cancel } handle - Centralized state cleanup via cleanupAgentState() - Credential handler consolidation (prepareProcessEnv) Prompts refactor: - Split monolithic prompts.ts into per-mode modules - Add workspace layout section to agent prompts - Fix markdown-to-tiptap double-serialization Server/tRPC: - Subscription heartbeat (30s) and bounded queue (1000 max) - Phase CRUD: approvePhase, deletePhase, dependency queries - Page: findByIds, getPageUpdatedAtMap - Wire new repositories through container and context	2026-02-09 22:33:28 +01:00
Lukas May	43e2c8b0ba	fix(agent): Eliminate race condition in completion handling PROBLEM: - Agents completing with questions were incorrectly marked as "crashed" - Race condition: polling handler AND crash handler both called handleCompletion() - Caused database corruption and lost pending questions SOLUTION: - Add completion mutex in OutputHandler to prevent concurrent processing - Remove duplicate completion call from crash handler - Only one handler executes completion logic per agent TESTING: - Added mutex-completion.test.ts with 4 test cases - Verified mutex prevents concurrent access - Verified lock cleanup on exceptions - Verified different agents can process concurrently FIXES: residential-cuckoo and 12+ other agents stuck in crashed state	2026-02-08 15:51:32 +01:00

Author

SHA1

Message

Date

Lukas May

fab7706f5c

feat: Phase schema refactor, agent lifecycle module, and log chunks

Phase model changes:
- Drop `number` column (ordering now by createdAt + dependency DAG)
- Replace `description` (plain text) with `content` (Tiptap JSON)
- Add `approved` status as dispatch gate
- Add phase dependency management (list, remove, dependents)
- Approval gate in PhaseDispatchManager.queuePhase()

Agent log chunks:
- New `agent_log_chunks` table for DB-first output persistence
- LogChunkRepository port + DrizzleLogChunkRepository adapter
- FileTailer onRawContent callback streams chunks to DB
- getAgentOutput reads from DB first, falls back to file

Agent lifecycle module (src/agent/lifecycle/):
- SignalManager: atomic signal.json read/write/wait operations
- RetryPolicy: exponential backoff with error-specific strategies
- ErrorAnalyzer: pattern-based error classification
- CleanupStrategy: debug archival vs production cleanup
- AgentLifecycleController: orchestrates retry/recovery flow
- Missing signal recovery with instruction injection

Completion detection fixes:
- Read signal.json file instead of parsing stdout as JSON
- Cancellable pollForCompletion with { cancel } handle
- Centralized state cleanup via cleanupAgentState()
- Credential handler consolidation (prepareProcessEnv)

Prompts refactor:
- Split monolithic prompts.ts into per-mode modules
- Add workspace layout section to agent prompts
- Fix markdown-to-tiptap double-serialization

Server/tRPC:
- Subscription heartbeat (30s) and bounded queue (1000 max)
- Phase CRUD: approvePhase, deletePhase, dependency queries
- Page: findByIds, getPageUpdatedAtMap
- Wire new repositories through container and context

2026-02-09 22:33:28 +01:00

Lukas May

43e2c8b0ba

fix(agent): Eliminate race condition in completion handling

PROBLEM:
- Agents completing with questions were incorrectly marked as "crashed"
- Race condition: polling handler AND crash handler both called handleCompletion()
- Caused database corruption and lost pending questions

SOLUTION:
- Add completion mutex in OutputHandler to prevent concurrent processing
- Remove duplicate completion call from crash handler
- Only one handler executes completion logic per agent

TESTING:
- Added mutex-completion.test.ts with 4 test cases
- Verified mutex prevents concurrent access
- Verified lock cleanup on exceptions
- Verified different agents can process concurrently

FIXES: residential-cuckoo and 12+ other agents stuck in crashed state

2026-02-08 15:51:32 +01:00

2 Commits