Codex Orange Book
- Source
- huasheng.ai
- Author
- 花叔
- Date
Context
The independent Codex Orange Book gives a concrete answer to repeated project corrections: record the working rules in AGENTS.md, the instruction file that OpenAI's Codex reads when it starts. The book's chapter on project instructions recommends recording build and test commands, local pitfalls, review requirements, and what counts as finished work.
The book also describes how those instructions combine. Codex gathers a global file and then files from the project root down to the current directory, with rules closer to the working directory taking priority. This lets a repository carry its own expectations into a new session. The book's test for what to include is whether the could work it out from the code: if not, write it down.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
“这本书覆盖 Codex 的五种形态（CLI / 桌面 App / Cloud / IDE 扩展 / Chrome 扩展），从安装到独立构建产品，并诚实对比 Claude Code 给出双线开发者的心智模型。”
“粗糙之处也没回避：1M 上下文实际可用 258K、Windows Full Access 真的删过全盘、MultiAgentV2 仍有未修 bug。”
Checking sign-in…
Loading comments…