Copilot instructions

1# GitHub Copilot Instructions — Second Brain
2 
3## Project
4 
5Developer knowledge graph — pnpm monorepo with Turborepo. Node.js 22+, TypeScript 5.8+ strict mode, ESM only.
6 
7For full architecture details, see [docs/architecture.md](../docs/architecture.md).
8For agent-level instructions, see [AGENTS.md](../AGENTS.md).
9 
10## Tech Stack
11 
12- **Runtime**: Node.js 22+, ESM modules only (`"type": "module"`)
13- **Language**: TypeScript 5.8+ (strict mode, explicit return types, no `any`)
14- **Database**: SQLite via better-sqlite3 + Drizzle ORM, WAL mode
15- **Search**: FTS5 (BM25) + sqlite-vec (cosine KNN), fused via Reciprocal Rank Fusion
16- **Validation**: Zod v4 for all external inputs
17- **Testing**: Vitest with in-memory SQLite (`:memory:`), globals enabled
18- **Build**: tsdown (ESM bundler), tsc (type-check only)
19- **Frontend**: React 19, Zustand, Cytoscape.js, Tailwind CSS, Radix UI
20- **Sync**: Yjs CRDTs via Hocuspocus relay
21- **IDs**: ULIDs (via ulidx). Never UUID
22- **Timestamps**: ISO 8601 strings. Never unix timestamps
23 
24## Conventions
25 
26### DO
27- Use named exports, re-export from package index.ts
28- Use workspace imports: `@second-brain/core`, `@second-brain/types`
29- Use Zod schemas for request validation in routes
30- Write tests with Vitest using in-memory SQLite
31- Use `batchUpsert` for bulk operations (merges observations/tags on conflict)
32- Keep observations as atomic, concise facts (string arrays)
33- Use try-catch with descriptive error messages
34 
35### DON'T
36- Use path aliases (rely on workspace resolution)
37- Use `any` type
38- Import from package internals (use public API via index.ts)
39- Use UUIDs (use ULIDs from ulidx)
40- Use Date objects or unix timestamps in storage
41- Manually write to entities_fts table (auto-maintained via triggers)
42- Sync the 'personal' namespace
43- Mutate stored confidence values (decay is read-side only)
44 
45## Monorepo Structure
46 
47```
48packages/types      — Shared TypeScript types
49packages/core       — Knowledge graph engine (Brain class)
50packages/collectors — Data collectors + streaming providers
51packages/ingestion  — LLM extraction + embedding pipeline
52packages/sync       — Yjs CRDT sync bridge
53packages/mcp-server — MCP tools (32 tools, stdio + HTTP)
54apps/server         — Express 5 REST API + WebSocket (port 7430)
55apps/ui             — React 19 + Cytoscape.js web app (port 5173)
56apps/relay          — Hocuspocus CRDT relay (port 7421)
57tools/cli           — brain CLI (Commander.js)
58```
59 
60## Commands
61 
62```bash
63pnpm install          # Install dependencies
64pnpm build            # Build all packages
65pnpm test             # Run all tests
66pnpm check-types      # Type-check everything
67pnpm dev              # Start dev servers
68```
69 
70## Data Model
71 
72- **Entity types** (15): concept, decision, pattern, person, file, symbol, event, tool, fact, conversation, reference, implementation, pull_request, merge_request, branch
73- **Relation types** (20): relates_to, depends_on, implements, supersedes, contradicts, derived_from, authored_by, decided_in, uses, tests, contains, co_changes_with, preceded_by, blocks, reviewed_by, merged_in_mr, merged_in_pr, touches_file, owns, parallel_with
74- Every entity/relation has a `namespace` field
75- Bitemporal: eventTime + ingestTime
76- Confidence with time-based decay (read-side only)
77