Jason Woltje
1aa11c4ee8
Implement @mosaic/brain — typed structured data service with MCP + REST API, JSON file backend, and schema validation via Zod. Collections: tasks, projects, events, agents, tickets, appreciations, missions, mission_tasks. MCP tools: brain_tasks, brain_projects, brain_events, brain_agents, brain_tickets, brain_today, brain_stale, brain_stats, brain_search, brain_audit, brain_missions, brain_mission, brain_mission_tasks, plus mutation tools for all collections. REST API mirrors MCP 1:1 at /v1/*. Bearer token auth with timing-safe comparison. Fastify server with per-request MCP instances (stateless HTTP transport). JSON file storage with proper-lockfile for concurrent access. Also adds Brain* types to @mosaic/types. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1.4 KiB
1.4 KiB
Scratchpad: @mosaic/brain Structured Data Service
Issue: #10 Branch: feat/brain-service PRD: jarvis-brain/docs/planning/MOSAIC-BRAIN-MEMORY-PRD.md
Objective
Implement @mosaic/brain as a Fastify service exposing structured data (tasks, projects, events, missions, etc.) via MCP + REST with JSON file backend.
Plan
- Add brain data types to @mosaic/types
- Scaffold packages/brain/ with Fastify + MCP SDK
- Implement JSON file storage adapter with locking
- Implement collections: tasks, projects, events, agents, tickets, appreciations, missions, mission_tasks
- Implement MCP tools (query + mutation + mission)
- Implement REST API (mirrors MCP)
- Bearer token auth middleware
- brain_today computed endpoint
- brain_stale, brain_stats, brain_audit, brain_search, brain_graph
- Dockerfile + docker-compose
- Tests
- Code review
- Commit, push, PR, merge
Progress
- 2026-03-10: Issue #10 created, branch created, starting implementation
Risks
- Types package already has
Task/TaskStatus/TaskPriorityfor queue — need namespacing - Large implementation surface — focus on core CRUD first, computed endpoints second
Decisions
- Brain types exported from
@mosaic/typesunderBrainprefix to avoid queue type conflicts - JSON storage uses one file per collection
- In-memory index rebuilt on startup