Covers today's quality + infra pass: brand/structural cleanup, MCP per-domain refactor (1697 -> 89 lines, 87 tools), canvas ConfirmDialog unification, 4 platform handler decompositions (+47 Go tests), E2E hardening for Phase 30.1/30.6 auth, and two new CI jobs (e2e-api + shellcheck). - CLAUDE.md: updated test counts (Go 536, canvas 357, SDK 121, MCP 97, workspace 1084); documented MCP per-domain split + new api.ts; added handler-decomposition section; Phase 30.1/30.6 auth callout; new CI jobs; env vars cross-ref. - PLAN.md: Phase 31 "Quality + Infra Pass" marked shipped; test totals refreshed to 2,295. - README.zh-CN.md: license badge MIT -> BSL 1.1; added BSL license block. - docs/api-protocol/platform-api.md: registry table gains Auth column documenting Phase 30.1 bearer-token and Phase 30.6 X-Workspace-ID requirements on heartbeat/update-card/discover/peers. - docs/development/local-development.md: updated stale test counts; added e2e-api + shellcheck CI jobs; pointer to new testing-e2e.md. - docs/development/testing-e2e.md: new — per-script reference, auth prerequisites, local run, CI coverage, adding-a-new-check checklist. - docs/edit-history/2026-04-13.md: top-of-file summary section added spanning PRs #1-#8; preserves existing per-feature entries below. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
5.1 KiB
Local Development
Starting the Stack
docker compose up
This starts:
| Service | Port | Description |
|---|---|---|
| Postgres | internal only | Primary database |
| Redis | internal only | Ephemeral state |
| Platform (Go) | :8080 |
Control plane API |
| Canvas (Next.js) | :3000 |
Visual frontend |
| Langfuse web | :3001 (host) / :3000 (internal) |
Observability UI |
| Langfuse worker | — | Background processing |
| ClickHouse | — | Langfuse dependency |
Each workspace container is provisioned on demand by the platform when a user creates or imports one.
Langfuse uses a dedicated langfuse Postgres database. The compose stack creates it automatically before starting the Langfuse service, so it does not conflict with the platform's molecule schema.
Infrastructure Only
To start just Postgres, Redis, and Langfuse (no application code):
docker compose -f docker-compose.infra.yml up
Optional Profiles
docker compose --profile multi-provider up # Add LiteLLM proxy (unified LLM API)
docker compose --profile local-models up # Add Ollama (local LLM models)
Environment Variables
Platform (Go)
DATABASE_URL=postgres://dev:dev@postgres:5432/molecule?sslmode=prefer
REDIS_URL=redis://redis:6379
PORT=8080
SECRETS_ENCRYPTION_KEY=dev-key-change-in-production
WORKSPACE_DIR=/path/to/molecule-monorepo # Optional global fallback; prefer per-workspace workspace_dir in org.yaml or API
Canvas (Next.js)
NEXT_PUBLIC_PLATFORM_URL=http://localhost:8080
NEXT_PUBLIC_WS_URL=ws://localhost:8080/ws
Workspace Runtime
WORKSPACE_ID= # assigned by platform on provision
WORKSPACE_CONFIG_PATH= # path to config folder inside container
MODEL_PROVIDER= # e.g. anthropic:claude-sonnet-4-6
TIER= # 1, 2, 3, or 4
PLATFORM_URL= # http://platform:8080
PARENT_ID= # set by platform during team expansion (empty for top-level)
ANTHROPIC_API_KEY= # or OPENAI_API_KEY, etc.
LANGFUSE_HOST= # http://langfuse-web:3000 (internal container port; host-mapped to :3001)
LANGFUSE_PUBLIC_KEY=
LANGFUSE_SECRET_KEY=
LANGSMITH_TRACING=true # LangGraph reads this to enable tracing
Technology Versions
Go 1.25+ (go.mod)
Python 3.11+
Node.js 22+
Next.js 15
React Flow 12 (@xyflow/react)
a2a-sdk 0.3+ (A2A server SDK, install with a2a-sdk[http-server])
langfuse 3.x (self-hosted Docker)
Postgres 16
Redis 7
Docker Compose 2.x
Running Tests
Unit Tests
cd platform && go test -race ./... # Go tests with race detection (536 tests)
cd canvas && npm test # Vitest tests (357 tests)
cd workspace-template && python -m pytest -v # Workspace runtime tests (1084 tests)
cd sdk/python && python -m pytest -v # SDK tests (121 tests)
cd mcp-server && npm test # MCP server tests (97 Jest tests)
Integration Tests
bash tests/e2e/test_api.sh # 62 API tests (quick local verify; Phase 30.1 bearer-auth aware; also runs in CI)
bash tests/e2e/test_a2a_e2e.sh # 22 A2A e2e tests (requires platform + 2 agents)
bash tests/e2e/test_activity_e2e.sh # 25 activity/task E2E tests (requires platform + 1 agent)
bash tests/e2e/test_comprehensive_e2e.sh # 67 endpoint/memory/bundle/approval checks
All E2E scripts share tests/e2e/_lib.sh helpers and are shellcheck-clean (enforced in CI). See ./testing-e2e.md for auth prerequisites and what CI runs.
CI Pipeline
GitHub Actions runs automatically on push to main and on PRs (.github/workflows/ci.yml):
- platform-build — Go build, vet,
go test -racewith coverage profiling (25% baseline threshold; setup-go uses module cache) - canvas-build — npm build,
vitest run(no--passWithNoTests-- tests must exist and pass) - mcp-server-build — npm build
- python-lint —
pytest --cov=. --cov-report=term-missing(pytest-cov enabled) - e2e-api (added 2026-04-13) — Postgres + Redis service containers, migrations applied via
docker exec,tests/e2e/test_api.shmust pass 62/62 - shellcheck (added 2026-04-13) — lints every
tests/e2e/*.sh
Postgres and Redis are not exposed to the host -- use docker compose exec postgres psql or docker compose exec redis redis-cli for direct access.
Utility Scripts
| Script | Purpose |
|---|---|
infra/scripts/setup.sh |
Initialize the local environment |
infra/scripts/nuke.sh |
Tear down and clean up everything |
bundle-compile.sh |
Compile workspace config folders into .bundle.json files |
test_api.sh |
Run 62 platform API integration tests |
test_a2a_e2e.sh |
Run 22 A2A end-to-end tests |
test_activity_e2e.sh |
Run 25 activity/task E2E tests |
setup-org.sh |
Create default 15-agent org hierarchy (PM + Marketing/Research/Dev teams, all Claude Code) |
Related Docs
- Architecture — System overview
- Observability — Langfuse details