# ChatGPT Handoff Rules

## Purpose

When the project owner moves from Codex back to ChatGPT or CCRA for discussion, Codex must provide a compact handoff document.

The handoff prevents ChatGPT from reconstructing project state from scattered history and prevents the owner from manually explaining the latest local implementation.

## When To Create A Handoff

Create a ChatGPT handoff when:

- The owner says they are going back to ChatGPT.
- A model extraction or repair session completes.
- Rules, schemas, workflow, indexes, or validation tools materially change.
- The next step requires product, CCRA, or architecture judgment outside Codex.

Do not replace validation reports with a handoff. The handoff summarizes and points to validated artifacts.

## Location And Naming

Store handoffs in `reports/`.

Use this naming pattern:

```text
ChatGPT交接文档_<topic>_<YYYY-MM-DD>.md
```

Example:

```text
reports/ChatGPT交接文档_模型库MVP_2026-06-16.md
```

## Required Sections

Each handoff must include:

1. 当前阶段
2. 本轮目标
3. 已完成内容
4. 关键文件路径
5. Codex 实现和原计划的差异
6. 当前问题
7. 需要 ChatGPT / CCRA 判断的事项
8. 下一步候选方向
9. 验证结果

## Required Distinctions

The handoff must distinguish:

- 已验证事实
- 当前草稿
- 待产品判断
- 待 CCRA / ChatGPT 判断
- 已知限制

Do not write only "已完成". Include the verification commands and results.

## Required File References

The handoff should point to local files rather than copying everything.

For model library MVP handoffs, include links to:

- `reports/validation_report.md`
- `models/model_index.json`
- `cards/card_index.md`
- `schemas/model_card.schema.json`
- current model JSON files
- current Markdown cards
- source and regression indexes
- relevant rules or workflow docs

## Relationship To Knowledge Assets

Handoffs are session-level reports.

Long-term reusable knowledge belongs in `knowledge_assets/`.

If a handoff repeats stable rules, model maps, or project context that should be reused across sessions, create or update a knowledge asset instead of leaving it only in `reports/`.