Original Summary

本帖使用社区开源推广，符合推广要求。我申明并遵循社区要求的以下内容： 我的帖子已经打上 开源推广 标签： 是 我的开源项目完整开源，无未开源部分： 是 我的开源项目已链接认可 LINUX DO 社区： 是 我帖子内的项目介绍，AI生成、润色内容部分已截图发出： 是 以上选择我承诺是永久有效的，接受社区和佬友监督： 是 以下为项目介绍正文内容，AI生成、润色内容已使用截图方式发出 github.com GitHub - ShinyNito/codex-review 通过在 GitHub 上创建帐户来为 ShinyNito/codex-review 开发做出贡献。 相信最近OpenAI过度拉胯导致应该有一堆佬们都逃到了Claude. 但是GPT这模型用来审查还是可以的. 加上最近Claude code 上了mod功能. 可以更加方便的进行拓展. 所以我就花了半小时vibe了一个小mod 用法非常简单就是 /codex-review [–read-only] [你的额外关注点] 切换审查模型有两种方式，没设置时用 Codex 自己的配置： 临时切换：/codex-review --model <模型名> --effort high，只对这一次审查生效。 持久设置：在 /config 的 codex-reviewer 下填 model，值会传给 codex exec -m，对之后每次审查生效。 都不设：用 ~/.codex/config.toml（或 $CODEX_HOME/config.toml）里顶层的 model 和 model_reasoning_effort。 优先级：命令参数 > /config > config.toml。推理强度没有 /config 选项，只能用 --effort 或 config.toml。状态条显示的是这次实际使用的模型和强度。 内置提示词 Review and fix the uncommitted changes in this workspace (visible in git status and git diff). Goal: bring these changes to a state a human can review as-is, written in a standard, readable style that would fit a mature open-source project. Focus: start with performance; remove duplicated validation, redundant state and duplicated tests; use modern idioms and do not hand-write what an imported package already provides; remove code that only forwards to something else; follow the conventions and layout of this project and of well-known open-source projects of the same kind. Scope: concentrate on these changes. Read surrounding code as needed to judge them, but only modify what relates to the changes. Done when: the problems you found are fixed directly, and the project's existing checks (tests, lint, type check) pass with no regressions. You may edit related files, run those checks and fix failures without asking first. Leave anything that needs a product or architecture decision unchanged and note it in the report. Finish with a short list: what you changed, why, and what is left unresolved. --read-only 模式（只报告） Review the uncommitted changes in this workspace (visible in git status and git diff). This is a read-only review: do not modify, create or delete any file. Goal: tell the author what stands between these changes and a state a human can review as-is, written in a standard, readable style that would fit a mature open-source project. Focus: start with performance; remove duplicated validation, redundant state and duplicated tests; use modern idioms and do not hand-write what an imported package already provides; remove code that only forwards to something else; follow the conventions and layout of this project and of well-known open-source projects of the same kind. Scope: concentrate on these changes. Read surrounding code as needed to judge them, and run the project's existing checks (tests, lint, type check) if they do not write to the workspace. Done when: every problem you found is listed with its file and line, why it matters and the fix you would make, most important first. Finish with that list, then a one-line verdict: ready to merge, or what blocks it. 1 个帖子 - 1 位参与者 阅读完整话题

  • 情报分类：商业与市场研究
  • 分类依据：内容涉及商业、投资或市场动态
  • 信息来源：服务器 / LINUX DO - 最新话题
  • 发布时间：2026/10/4 15:30:21