关于
The commit-changes skill stages, commits, and amends Git changes using conventional commit messages. It enables selective staging, descriptive HEREDOC-formatted messages, and commit history verification. Use it to save logical work units to version control with proper commit practices.
快速安装
Claude Code
推荐npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/commit-changes在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
Commit Changes
Stage files selective. Write clear commit messages. Verify commit history.
When Use
- Save logical unit of work to version control
- Create commit with descriptive, conventional message
- Amend most recent commit (message or content)
- Review what commit before commit
Inputs
- Required: One or more changed files to commit
- Optional: Commit message (drafted if not given)
- Optional: Whether amend previous commit
- Optional: Co-author attribution
Steps
Step 1: Review Current Changes
Check working tree status. Inspect diffs.
# See which files are modified, staged, or untracked
git status
# See unstaged changes
git diff
# See staged changes
git diff --staged
Got: Clear picture of all modified, staged, untracked files.
If fail: git status fail? Verify inside git repo (git rev-parse --is-inside-work-tree).
Step 2: Stage Files Selective
Stage specific files. Avoid git add . or git add -A — may grab sensitive files or unrelated changes.
# Stage specific files by name
git add src/feature.R tests/test-feature.R
# Stage all changes in a specific directory
git add src/
# Stage parts of a file interactively (not supported in non-interactive contexts)
# git add -p filename
Review staged before commit.
git diff --staged
Got: Only intended files and changes staged. No .env, credentials, large binaries.
If fail: Unstage accidental files with git reset HEAD <file>. Sensitive data staged? Unstage now, before commit.
Step 3: Write Commit Message
Use conventional commits format. Always pass message via HEREDOC for proper formatting.
git commit -m "$(cat <<'EOF'
feat: add weighted mean calculation
Implements weighted_mean() with support for NA handling and
zero-weight filtering. Includes input validation for mismatched
vector lengths.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
EOF
)"
Conventional commit types:
| Type | When use |
|---|---|
feat | New feature |
fix | Bug fix |
docs | Documentation only |
test | Add or update tests |
refactor | Code change — no fix, no new feature |
chore | Build, CI, dependency updates |
style | Formatting, whitespace (no logic change) |
Got: Commit created with descriptive message. Explains why, not just what.
If fail: Pre-commit hook fail? Fix issue, re-stage with git add, create new commit. No --amend — failed commit never created.
Step 4: Amend Last Commit (Optional)
Only amend if commit not pushed to shared remote.
# Amend message only
git commit --amend -m "$(cat <<'EOF'
fix: correct weighted mean edge case for empty vectors
EOF
)"
# Amend with additional staged changes
git add forgotten-file.R
git commit --amend --no-edit
Got: Previous commit updated in-place. git log -1 shows amended content.
If fail: Commit already pushed? No amend. Create new commit. Force-push amended commits to shared branches breaks history.
Step 5: Verify Commit
# View the last commit
git log -1 --stat
# View recent commit history
git log --oneline -5
# Verify the commit content
git show HEAD
Got: Commit appears in history with correct message, author, file changes.
If fail: Commit has wrong files? Use git reset --soft HEAD~1 to undo commit, keep changes staged. Re-commit correct.
Checks
- Only intended files in commit
- No sensitive data (tokens, passwords,
.envfiles) committed - Commit message follows conventional commits format
- Message body explains why change made
-
git logshows commit with correct metadata - Pre-commit hooks (if any) passed
Pitfalls
- Commit too much at once: Each commit = one logical change. Split unrelated changes into separate commits.
- Using
git add .blind: Always reviewgit statusfirst. Prefer stage specific files by name. - Amend pushed commits: Never amend commits pushed to shared branch. Rewrites history. Breaks collaborators.
- Vague commit messages: "fix bug" or "update" tells nothing. Describe what changed and why.
- Forget
--no-editon content amends: Adding forgotten files to last commit? Use--no-edit— keeps existing message. - Hook fail leading to
--amend: Pre-commit hook fail → commit never created.--amendwould modify previous commit. Always create new commit after fix hook issues.
See Also
manage-git-branches- branch workflow before commitcreate-pull-request- next step after commitresolve-git-conflicts- handle conflicts during merge/rebaseconfigure-git-repository- repo setup and conventions
GitHub 仓库
Frequently asked questions
What is the commit-changes skill?
commit-changes is a Claude Skill by pjt222. Skills package instructions and resources that Claude loads on demand, so Claude can perform commit-changes-related tasks without extra prompting.
How do I install commit-changes?
Use the install commands on this page: add commit-changes to Claude Code as a plugin, or clone its repository into your skills directory, then restart Claude so it picks up the skill.
What category does commit-changes belong to?
commit-changes is in the Meta category, tagged general.
Is commit-changes free to use?
Yes. commit-changes is listed on AIMCP and free to install. It runs inside Claude, so no separate service account is required to use the skill itself.
相关推荐技能
Content Collections 是一个 TypeScript 优先的构建工具,可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计,提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。
这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持,涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流,可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。
该Skill帮助开发者创建OpenCode插件,用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式,适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。
SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。
