Git变更检查
git-dirty-check
by amitb-quantum
Read-only triage for a local git working tree that summarizes uncommitted changes and applies conservative risk flags. Use when a user asks what changed in a repo, what is dirty, what is staged versus unstaged, or which risk flags apply before a commit. Do not use for commit writing, branch management, history analysis, merge conflict resolution, PR review, or general git assistance.
安装
claude skill add --url https://github.com/openclaw/skills文档
git-dirty-check
A narrow git skill for summarizing uncommitted changes in a local repository. It groups staged, unstaged, untracked, and conflicted entries, applies conservative path- and filename-based risk flags, and suggests up to three next checks without modifying repo state.
git-dirty-check is a deliberately narrow skill for one job: understanding the current uncommitted state of a local git repository.
It does not try to manage branches, write commit messages, resolve conflicts, review pull requests, or act as a general git assistant. Instead, it gives a compact, structured triage of the current working tree:
- repo state
- changed entries by category
- conservative risk flags based mostly on paths, filenames, and diff stats
- a few copy-paste-ready next checks
- explicit omissions when details are capped or sensitive
The skill is read-only. It is designed for developers who want a quick operational summary before a commit, handoff, or local review, without granting write access or triggering broader repo assistance.
It also applies strict limits to deeper inspection:
- fail fast if the target is not a git repo
- branch metadata is optional
- deep diff inspection is capped
- secret-bearing filename patterns are handled at filename level only
This skill is most useful when raw git output is technically sufficient, but slower to parse than a stable triage summary.
Workflow
- Confirm the target path is inside a git repository.
- If it is not a git repository, stop immediately and say so.
- Collect read-only repo state with git status and diff-stat commands.
- Group changed entries into:
- staged
- unstaged
- untracked
- conflicted
- Apply conservative risk flags based mostly on file paths and filenames.
- Inspect diff content only when filename and diff-stat data are insufficient to assign a listed risk flag, and only for files not matched by the secret-bearing file rule.
- Return the output in the fixed order defined below.
Commands to Prefer
Use read-only commands only.
Preferred commands:
git rev-parse --show-toplevelgit rev-parse --abbrev-ref HEADgit status --shortgit diff --statgit diff --cached --stat
Use targeted diff inspection only under the deep diff inspection cap, and only for files not matched by the secret-bearing file rule.
Fail-Fast Rule
If the target path is not inside a git repository:
- stop immediately
- report that the path is not a git repository
- do not continue with generic filesystem analysis
Branch Handling
Branch name is optional metadata only. If branch name is available from a preferred read-only command, include it in Repo state. If HEAD is unavailable or branch name lookup fails in an otherwise valid repo, omit branch metadata without error. Do not give branch advice.
Deep Diff Inspection Cap
Deep diff inspection is optional and must remain narrow.
Hard limits:
- inspect at most 3 files
- inspect at most 40 lines per file
- never inspect secret-bearing files beyond filename-level detection
Use deep diff inspection only for files not matched by the secret-bearing file rule and only when filename and diff-stat data are insufficient to assign or decline a listed risk flag.
Secret-Bearing File Rule
Treat files matching these secret-bearing filename patterns as filename-level only.
Examples:
.env.env.**.pem*.keyid_rsaid_ed25519secrets.*
For these files:
- do not inspect diff content
- do not print values
- report only that an entry matched a secret-bearing filename pattern
Conservative Risk Heuristics
Keep risk flags conservative and mostly path/file based.
Flag examples:
- dependency manifest changed
- lockfile changed
- manifest changed without lockfile
- lockfile changed without manifest
- CI or workflow file changed
- possible deploy or infrastructure file changed
- possible auth or security-related file changed
- possible migration or schema file changed
- more than 10 unique paths changed
- entry matched a secret-bearing filename pattern
Do not over-interpret. Avoid speculative impact claims.
Output Order
Always return sections in this exact order:
- Repo state
- Changed files by category
- Risk flags
- Suggested next checks
- Unknowns or omitted detail
Output Guidance
Keep the response brief and structured.
1. Repo state
Include:
- repo root
- branch name only as optional metadata
- counts of staged, unstaged, untracked, and conflicted files
2. Changed files by category
List entries under:
- staged
- unstaged
- untracked
- conflicted
List entries directly when there are 10 or fewer in a category. Otherwise list the first 10 and summarize the remainder.
3. Risk flags
List only conservative flags supported by filenames, paths, diff stats, or tightly capped non-sensitive diff samples.
4. Suggested next checks
Give 1 to 3 copy-paste-ready commands. Prefer read-only commands.
5. Unknowns or omitted detail
Say what was intentionally not inspected, such as:
- diff content omitted due to caps
- sensitive files not inspected beyond filename
- large repo details summarized only
Boundaries
Never:
- modify git state
- stage, unstage, commit, stash, reset, checkout, merge, rebase, or push
- write commit messages
- provide branch strategy
- expand into PR review or general git help
- expose secret-looking values from diffs
- use network operations
Stay narrow. This skill exists to answer: "What changed here, and which conservative risk flags apply?"
Trigger examples
- “What changed in this repo?”
- “Summarize my uncommitted changes.”
- “What’s dirty here?”
- “Give me a quick git change triage.”
- “Before I commit, what changed and which risk flags apply?”
- “Show staged vs unstaged vs untracked changes in this repo.”
Anti-trigger examples
- “Write a commit message.”
- “Help me resolve this merge conflict.”
- “Review this PR.”
- “Undo my last commit.”
- “Clean up my branch.”
- “Find the bug in this repo.”
- “Refactor these files.”
- “Explain git rebase.”
Limitations
- Read-only only, no git state changes
- Focused on current uncommitted state, not history
- Risk flags are conservative and mostly path/file based
- Secret detection uses filename patterns, not full secret scanning
- Deep diff inspection is tightly capped
- Summary usefulness depends on the clarity of repo file naming and layout
- Not a replacement for full code review or git expertise
Safety notes
- Uses read-only git commands only
- Does not stage, unstage, commit, stash, reset, checkout, merge, rebase, or push
- Does not use network operations
- Fails fast on non-repo paths
- Does not inspect secret-bearing files beyond filename-level detection
- Avoids exposing secret-looking values from diffs
- Keeps output scoped to triage, not general repo management
相关 Skills
前端设计
by anthropics
面向组件、页面、海报和 Web 应用开发,按鲜明视觉方向生成可直接落地的前端代码与高质感 UI,适合做 landing page、Dashboard 或美化现有界面,避开千篇一律的 AI 审美。
✎ 想把页面做得既能上线又有设计感,就用前端设计:组件到整站都能产出,难得的是能避开千篇一律的 AI 味。
网页应用测试
by anthropics
用 Playwright 为本地 Web 应用编写自动化测试,支持启动开发服务器、校验前端交互、排查 UI 异常、抓取截图与浏览器日志,适合调试动态页面和回归验证。
✎ 借助 Playwright 一站式验证本地 Web 应用前端功能,调 UI 时还能同步查看日志和截图,定位问题更快。
网页构建器
by anthropics
面向复杂 claude.ai HTML artifact 开发,快速初始化 React + Tailwind CSS + shadcn/ui 项目并打包为单文件 HTML,适合需要状态管理、路由或多组件交互的页面。
✎ 在 claude.ai 里做复杂网页 Artifact 很省心,多组件、状态和路由都能顺手搭起来,React、Tailwind 与 shadcn/ui 组合效率高、成品也更精致。
相关 MCP 服务
GitHub
编辑精选by GitHub
GitHub 是 MCP 官方参考服务器,让 Claude 直接读写你的代码仓库和 Issues。
✎ 这个参考服务器解决了开发者想让 AI 安全访问 GitHub 数据的问题,适合需要自动化代码审查或 Issue 管理的团队。但注意它只是参考实现,生产环境得自己加固安全。
Context7 文档查询
编辑精选by Context7
Context7 是实时拉取最新文档和代码示例的智能助手,让你告别过时资料。
✎ 它能解决开发者查找文档时信息滞后的问题,特别适合快速上手新库或跟进更新。不过,依赖外部源可能导致偶尔的数据延迟,建议结合官方文档使用。
by tldraw
tldraw 是让 AI 助手直接在无限画布上绘图和协作的 MCP 服务器。
✎ 这解决了 AI 只能输出文本、无法视觉化协作的痛点——想象让 Claude 帮你画流程图或白板讨论。最适合需要快速原型设计或头脑风暴的开发者。不过,目前它只是个基础连接器,你得自己搭建画布应用才能发挥全部潜力。