免费在 GitHub 公开
check
Self-verification loop. After completing a task, re-read the original request, list each requirement, verify whether it was met, and fix gaps before stopping.
中文简介正在准备中。当前显示原文说明。
查看安装方法包含的文件（1）
- SKILL.md943 B
SKILL.md（原文）
安装前可以查看交给 AI 代理的指令内容。
check
Run after any non-trivial change. Catches drift where you implemented something adjacent to the request but not the actual request.
When to use
- Multi-step tasks (3+ requirements)
- Tasks where the user gave specific constraints (file paths, function names, edge cases)
- After tool failures or pivots — make sure the original goal still holds
How
- Quote the original request back, verbatim.
- For each verb / requirement, mark: done, partial, missing.
- For each partial or missing item, do the work now.
- Only after the list is all done, stop.
Never claim a task is complete based on intent. Only based on verified state — files that exist, tests that pass, output that was actually produced.
评价
还没有评价。欢迎分享你的使用感受。
评论
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。