Correct
从仓库历史找出 agent 反复犯的错误类别,逐类在能生效的最高层级让它不可能再犯,并证明新检查会在真实旧错误上失败。用于用户显式调用 /correct 或 $correct。
Install
npx skills add Dimon94/skills --skill correctgit clone --depth 1 https://github.com/Dimon94/skills.git /tmp/skills
mkdir -p ~/.claude/skills
cp -r /tmp/skills/skills/correct ~/.claude/skills/correctPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Correct 只在用户显式调用时执行。用户在这个仓库里反复纠正 agent 犯的同类错误。改仓库,让下一个 agent 犯不了。 假设每个贡献者都是 agent:只看自己打开的文件,照抄最近的例子,走能编译的最短路径。把仓库设计成“从一个文件看着对的改动,对整个仓库也对”。 用户给了错误描述时从它开始;没给时从历史里自己找。会话级的教训归 `retro`;本 Skill 只改仓库。 ## 1. 找错误类别 读最近的 commit、revert、review 评论、agent 指令文件,以及解释 workaround 的注释。把错误归成类。一类出现过两次才算数。 完成条件:每类都有至少两处证据坐标(commit SHA、评论链接或文件行号)。 ## 2. 逐类选能生效的最高层级 按 `principle-encode-lessons-in-structure` 判断,从高到低选: 1. **架构消除。** 每份状态一个 owner,每件事一种受支持的做法。隐藏内部实现,让错误的 import 失败。用单一真相源替换手工同步的清单。删掉 agent 会照抄的旧做法和死代码。 2. **类型强制,让坏状态写不出来。** 坏代码仍能编译时,加 lint 或 CI 检查,报错里写明该改用的文件、类型或函数。模式已经很普遍时,只在改动新增它时失败。 3. **测试行为。** 行为坏了仍通过的测试、只改内部就失败的测试,修掉或删掉。 4. **文档或 agent 规则放最后,只用于判断题。** agent 跳过它们时什么都不会失败。 完成条件:每类都选定一个层级,并写明更高层级为什么不行。 ## 3. 修并证明 先修出现最多的类别,一类一个独立改动。每个新检查都要在一个真实的旧错误上证明它会失败:把那次错误重放到当前代码上,看检查变红,再撤掉。本地和 CI 跑同一条命令。例外写在出问题的那一行,带理由、过期日期和人类批准。 改动在项目规定的任务树里做;提交按项目规则,用户明确要求时才执行。 完成条件:每个已修类别都有一次“旧错误让检查失败”的运行输出。 ## 4. 维护规则表 在项目的 agent 指令文件里维护一张表,每条规则对应强制它的东西。用户纠正你时,修掉错误并加上规则。规则已经在表里却没有东西强制它,就是一次重犯,在同一个改动里按第 2 步提到最高层级。错误已经不可能发生时,删掉那条规则。 ## 回复 逐类给出:证据、选定的层级、更高层级为什么不行、证明失败的那次运行。没修的类别列出来,说明原因。
Details
- Skill name
correct- Repository
- Dimon94/skills
- Path
skills/correct/SKILL.md- Repo updated
- 18 November 2026
FAQ
What does the Correct skill do?
从仓库历史找出 agent 反复犯的错误类别,逐类在能生效的最高层级让它不可能再犯,并证明新检查会在真实旧错误上失败。用于用户显式调用 /correct 或 $correct。
How do I install the Correct skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add Dimon94/skills --skill correct. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Correct popular?
It lives in Dimon94/skills, which has 107 GitHub stars and 21 forks.
Related skills
Maintain Verification Skill
107Dimon94/skills
Periodic pass that keeps a project's verification skill and feature map honest: parallel source readers per feature, one live session driving every feature, at most one PR of proven corrections. Use for /maintain-verification-skill or "audit the verify skill".
Create Verification Skill
107Dimon94/skills
Generate a project-local verification skill that drives your app the way a user does — any language, framework, or platform. Use for /create-verification-skill, "make a control skill for this repo", or when a project has no scripted way to prove UI/CLI/service behavior.
How
107Dimon94/skills
Use for "how does X work", code walkthroughs before changing something, and placement / ownership / layering questions ("where should this live", "which package owns this", "is this the right layer"). Explains subsystem architecture, runtime flow, onboarding mental models. Use why for motivation.
Flywheel
43aws-samples/sample-kiro-cli-multiagent-development
Analyze recent session transcripts for user-correction patterns and propose agent/steering/skill configuration improvements. Use when asked to run the flywheel, improve agent configuration, or learn from past sessions.
Aqara Agent
48aqara/aqara-agent-skills
aqara-agent is an official Aqara Home AI Agent skill. It supports natural-language home and space management, device inquiry, device control, ambience_create (氛围创建: apply or compose a named mood in a room or whole home via scene match then lights + optional music), ambience / mood orchestration reference, lighting effect inquiry and control (named lighting presets), device configuration (firmware/OTA upgrade only - not supported: rename devices, move devices between rooms or positions, or other device record edits outside firmware), device logs, scene management (query, execute, create, snapsh
Principle Explain the Number
107Dimon94/skills
Apply before you measure, and before you trust, report, or act on a number you measured: a speedup, a regression, a throughput, a latency, or an eval result. Find what limits it, and rule out that it measured something other than the work you think.
Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.