# CLAUDE.md

愈心谷 is an **AI-first** Monorepo. Treat `.ai/` as the executable engineering system.

## On every task

1. Read [AGENTS.md](AGENTS.md) (load order is authoritative).
2. Especially: `.ai/ai-contract.md`, `constitution.md`, `architecture.md`, `domain.md`, `forbidden.md`.
3. Prefer `patterns/` + `examples/` + `playbooks/` over free-form invention.
4. Finish with Review + DoD + matching checklist.

## Layout

| Path | Role |
|---|---|
| `.ai/` | AI Engineering System (source of truth for rules) |
| `apps/api` | Go API |
| `apps/user-h5` | Vue3 user H5 |
| `apps/mini-program` | Mini-program scaffold |
| `packages/*` | sdk / types / utils / ui |
| `apps/docs/` | Human PRD / business docs (not a substitute for `.ai/`) |

## Commands

```bash
export GOPROXY=https://goproxy.cn,direct   # if needed
cd apps/api && go run ./cmd/server
npm install && npm run dev:h5
```

## Legacy

Root `yuxingu.html`, `pages/`, `server.py` = prototype. Do not extend. See [LEGACY.md](LEGACY.md).

## Skill routing

When the user's request matches an available skill, invoke it. When in doubt, invoke the skill.

- Product ideas/brainstorming → `/office-hours`
- Strategy/scope → `/plan-ceo-review`
- Architecture → `/plan-eng-review`
- Design → `/design-consultation` or `/plan-design-review`
- Full review pipeline → `/autoplan`
- Bugs/errors → `/investigate`
- QA → `/qa` or `/qa-only`
- Code review → `/review`
- Ship/PR → `/ship` or `/land-and-deploy`
- Save/resume context → `/context-save` / `/context-restore`
- Spec/issue → `/spec`