feat: multi-stage Docker build + e2e test suite

[bupao]

Summary

• Multi-stage Dockerfile: TypeScript compiles inside Docker (builder stage), eliminating local Node.js requirement for image builds
• Native addon fix: Explicit npm rebuild better-sqlite3 after --ignore-scripts install to compile the .node binding for the target platform
• E2e test suite: scripts/e2e-test.sh — 26 automated test cases covering the full Docker workflow
• Build tooling: .dockerignore for faster context, Makefile/picoclaw.sh updated to remove local build dependencies

Changes

Dockerfile (multi-stage build)

• Stage 1 (builder): npm ci + npm run build — compiles TypeScript
• Stage 2 (runtime): npm ci --omit=dev --ignore-scripts + npm rebuild better-sqlite3 — production deps with native addon
• COPY --from=builder /build/dist/ ./dist/ — only compiled output enters runtime image

Build tooling

• Makefile: docker-build and docker-build-lambda no longer depend on build-ts
• picoclaw.sh: build_image() removed npm run build step
• .dockerignore: excludes node_modules, dist, dev-data, docs, .git

E2e test suite (scripts/e2e-test.sh)

26 test cases in 9 sections:

Section	Tests
Health check	GET /health returns status=ok
Authentication	No token → 401, wrong token → 401
Multi-turn conversation	4 rounds: establish identity → recall name → recall 3 facts → summarize
Conversation isolation	New conversation has no knowledge of previous
Task CRUD	Create → list → update → check → delete
Skills & memory	Skill sync verified, persona CLAUDE.md accessible
Container restart	Graceful stop → DB sync → restart → session resume with full context
Error handling	Non-existent conversation → 404
Graceful shutdown	POST /control/stop → container exits cleanly

Usage:

make test-e2e              # full: build + all tests (requires .env with API credentials)
make test-e2e-quick        # skip build and Claude API calls
./scripts/e2e-test.sh --no-build   # use existing image
./scripts/e2e-test.sh --no-chat    # skip tests that call Claude API

Documentation

• CLAUDE.md: Docker workflow section updated
• CHANGELOG.md: unreleased changes documented
• docs/SERVERLESS_API_DEPLOYMENT_GUIDE.md: multi-stage build docs, troubleshooting updated

Test results

── 1. Health Check ──
  PASS GET /health returns status=ok
── 2. Authentication ──
  PASS No token → 401
  PASS Wrong token → 401
── 3. Multi-Turn Conversation ──
  PASS Round 1: conversation created
  PASS Persona: agent identifies as TestBot
  PASS Round 2: agent recalls name 'Kevin'
  PASS Round 3: agent recalls all 3 facts (Kevin, Rust, Mochi)
  PASS Round 4: 4-turn conversation completed
  PASS Metadata: 8 messages, status=idle
── 4. Conversation Isolation ──
  PASS Isolation: new conversation_id
  PASS Isolation: agent reports no knowledge of Kevin
── 5. Task CRUD ──
  PASS Create/List/Update/Check/Delete
── 6. Skills & Memory ──
  PASS Skills synced, Persona accessible
── 7. Container Restart Persistence ──
  PASS Graceful stop → DB sync → restart → session resume (Mochi + Rust)
── 8. Error Handling ──
  PASS Non-existent conversation → 404
── 9. Graceful Shutdown ──
  PASS Container exited cleanly

Total: 26  Pass: 26  Fail: 0  Skip: 0

Test plan

• npm run build — TypeScript compiles cleanly
• npm test — 16 unit tests pass
• npm run format:check — Prettier formatting verified
• docker build --platform linux/amd64 — image builds successfully
• ./scripts/e2e-test.sh — 26/26 e2e tests pass in Docker container
• Documentation consistency across CLAUDE.md, CHANGELOG.md, deployment guide

🤖 Generated with Claude Code