OpenAI 路线更偏系统化：AGENTS.md 只做地图，真实知识进入 docs / plans，约束交给 lint / CI / structural tests，观测反馈直接喂给 agent。
适合怎么读
先看本专题首页，再按卡片进入各原始材料；若需要站内全景，可跳到站点地图与模板索引。
本专题包含
14 个原始条目，均保留源文件路径、站内相对链接与外部参考链接。
专题抓手
- Repo as source of truth
- Plan as artifact
- Mechanized invariants + observability + doc gardening
文档清单
AGENTS.md — OpenAI 风格的 Codex 工程 Harness（中文版）
这个文件应当 刻意保持简短 。 它是 地图 ，不是百科全书。
AGENTS.md — OpenAI-style harness for Codex-first engineering
This file is intentionally short. It is a map , not the manual.
Harness Engineering（OpenAI 版）：把 Codex 变成可持续交付的软件工程系统
这份目录基于 OpenAI 官方技术博客《工程技术：在智能体优先的世界中利用 Codex》： https://openai.com/zh Hans CN/index/harness engineering/
Framework 1 — Agent-First Product Delivery
Use this when building a new product or major product area primarily through Codex.
Framework 3 — Bug Reproduction / Fix / Regression Factory
Use this when the main workload is diagnosing and fixing bugs rapidly.
Framework 4 — Documentation Garden / Knowledge Reliability
Use this when the repository's main bottleneck is stale or fragmented knowledge.
Framework 2 — Long-Running Multi-Session Engineering
Use this when a task will span many agent sessions or many context windows.
OpenAI Harness Engineering — 读后感 / 总结 / 可迁移经验
也就是： 给什么文档 如何组织知识 如何定义边界 如何给验证闭环 如何给清理与治理机制
Architecture invariants
These are rules the repository should enforce mechanically where possible.
docs/index.md
This directory is the repository knowledge map.
Execution Plan: <title>
本页收录该主题的原始材料与整理内容。
feature_list.json
[ { "id": "feature 001", "category": "functional", "priority": "high", "description": "<One user visible end to end capability ", "steps": [ "<Open or start the system ", "<Perform the user action ", "<Observe …
!/usr/bin/env bash
echo "[init] repo: $(pwd)"
progress.md — shift handoff log
Append only. Never overwrite prior entries.