# AGENTS.md — AI System Entry

You are the software engineer for **愈心谷 (YuXinGu)**.

This file is for AI agents.

## Load order (every task)

1. `.ai/ai-contract.md`
2. `.ai/constitution.md`
3. `.ai/architecture.md`
4. `.ai/domain.md` (+ product/API tasks: `domain/domain-map.md`)
5. `.ai/forbidden.md`
6. `.ai/file-map.md` + `.ai/workflow.md`
7. Task rules: `.ai/coding.md` / `api.md` / `database.md` / `ui.md` / `security.md` / …
8. **Product scope:** `.ai/product/lexicon.md` + `feature-map.md` + `user-journey.md` + `page-tree.md`（冻结见 `ENGINEERING-FREEZE.md`）
9. **Any UI work:** `.ai/design/design-system.md` + `component-catalog.md` + matching `design/platform/*.md` (H5 default)
10. Matching **ADR** in `.ai/adr/` before changing stack or API style
11. Prefer **patterns/** + **examples/** + **playbooks/** over inventing structure
12. Before Done: `.ai/review.md` + `.ai/definition-of-done.md` + `.ai/checklists/*`
13. Verify via `.ai/commands.md`
14. `prompts/` are optional helpers — not a substitute for rules above

## Hard constraints

- Never violate architecture or Accepted ADRs.
- Never invent APIs, DB tables, or domain synonyms.
- Never guess requirements — **ASK FIRST**.
- Never modify files outside the current task.
- New code only under `apps/` and `packages/` (unless legacy migration task).
- One concern per change set. Function ≤50 lines. File ≤400 lines.
- Print the Review block when a coding task finishes.

## Cursor

`.cursor/` is an Agent Runtime view of this system via symlinks. Edit `.ai/` only.

## Growth

Do not invent Phase 2/3 folders (graph/mcp/skills/anti-patterns/…) unless `ROADMAP.md` says so and the task asks. Prefer improving patterns/examples/playbooks.

## Product one-liner

档案 → 性格探索/个人画像 → 关系理解 → 问答陪伴 → 会员/深度版 →（后）顾问。  
语言契约：`.ai/product/lexicon.md`。参考竞品模型，不用竞品品牌词与运势/解码叙事。