Five-dimension design quality review — score the artifact against craft, brand, accessibility, and copy, then fix what falls short before handing it over.
复制下面这句话,粘贴给 Claude Code、Codex、Cursor 等 AI 编程工具,它会读取安装说明并在你确认后完成安装。
请阅读 https://ai.atlankj.com/install/asset/gh-critique-theater-aa4357ff6043 ,按照其中的说明把「critique-theater」安装到你(当前 AI 工具)中。执行前先告诉我将运行的命令和写入的位置,等我确认。
查看 AI 将读取的安装说明正在读取 GitHub 原文…
内容来自 GitHub 原始文件,由原作者维护。在 GitHub 查看
Before you hand an artifact over, review it against five dimensions and fix what falls short. This is a quality bar, not a performance: do the review, apply the fixes, and describe the outcome in ordinary prose.
Judge the artifact you just produced from five angles. Each one owns its own scope; do not let a strength in one excuse a failure in another.
Record each concrete defect and fix it in the same turn. A finding you cannot fix is worth stating plainly, with the reason, so the user can decide.
Keep iterating while defects remain and you still have room to work. Stop when the artifact clears all five dimensions, or when further rounds stop producing real improvements — whichever comes first. Three passes is a sensible ceiling.
OD_MAX_DEVLOOP_ITERATIONS caps the outer plugin pipeline stage; it is not
this review's round limit. A pipeline's critique.score signal is a
scheduler-facing projection, not something you should write into your answer.
Write the outcome the way you would write any other part of your answer: plain prose, in the user's language, saying what you checked, what you fixed, and anything you deliberately left alone.
Do not invent a structured review transcript, a scored panel, a cast of named
reviewer personas, or any tag-based envelope around your answer. Do not write a
critique.json file into the project. None of those are inputs to anything —
they only add noise to the user's chat.