# Template: AGENTS.md

## Purpose
为 Codex 提供项目级执行约束，让实现、审查和交接都遵守同一套边界。

## When to Use
- Setup 阶段
- 项目需要同时支持 `Claude Code` 和 `Codex` 时
- 需要给 Codex 一个稳定的项目执行入口时

## Inputs
- 清洗后的会议纪要
- 技术轨道决策
- 项目目标与边界
- 已确认的 workflow / playbook

## Outputs
- `AGENTS.md`

## Primary Agent/Model
`GPT-5.4 Pro xhigh`

## Secondary Agent/Model
`Claude Opus 4.6`

## Required Skills
- `agent-md-writing`
- 若未安装，按 `claude-md-template.md` 的约束等价改写

## Steps
1. 用下方模板生成草案。
2. 把 `CLAUDE.md` 中的稳定规则压缩为 Codex 可直接执行的指令。
3. 明确 Read First 文件、禁止项、审批点和 handoff 方式。

## Exit Criteria
- Codex 在只读 `AGENTS.md` 的情况下，也能知道项目目标、技术边界、流程门禁和升级点

## Failure Recovery
- 若 `AGENTS.md` 只是重复愿景、不写边界与门禁，回到模板重写

## Related Templates
- [`claude-md-template.md`](./claude-md-template.md)
- [`agent-handoff-template.md`](./agent-handoff-template.md)

## Template
```markdown
# AGENTS.md

## Mission
- Project:
- Goal:
- Audience:
- In Scope:
- Out of Scope:

## Read First
- `CLAUDE.md`
- `ANALYSIS.md`
- `TODO.yaml`
- Selected playbook:
- Key specs:

## Workflow Contract
- Follow stages:
- Current playbook:
- Reset points:
- Human approval gates:

## Stack Guardrails
- Track:
- Language:
- UI:
- Styling:
- Auth:
- Data Fetching:
- Router:
- Service Layer:
- ORM:
- Database:

## Model Routing
- Primary planning and implementation:
- Secondary review:
- Fallback:

## Boundaries
- Do not introduce:
- Do not modify:
- Do not bypass:
- Ask before:

## Execution Defaults
- Preferred testing posture:
- Refactor rhythm:
- Handoff format:
- Done means:
```