commit Skill
Enforces a 17-step analysis and specific formatting rules for writing and reviewing high-quality, technical, and actionable git commit messages.
- GitHub Stars
- 13
- Dateien
- 1
- Letzter Commit
- 10. Nov. 2026
- Lizenz
- —
- Erlaubte Tools
- —
Was macht der commit Skill?
commit ist ein Agent Skill. Enforces a 17-step analysis and specific formatting rules for writing and reviewing high-quality, technical, and actionable git commit messages.
Zusammenfassung erzeugt aus der SKILL.md des Skills.
Über den commit skill
Developers who need to maintain high-quality, technical git histories can use this skill to enforce rigorous standards for commit messages. It functions as a pre-commit analysis engine that requires a 17-step review process before any message is written or reviewed. An agent should load this skill when a user requests commit requirements, guidelines, or specific commands like /ar:commit to ensure every message is actionable and technically precise.
The skill relies on executing git commands such as git status, git diff --staged, and git log to gather context and verify changes. It produces structured, self-documenting commit messages that include a concrete subject line, previous behavior descriptions, specific technical changes, and testable outcomes. Unlike standard commit practices, it strictly bans vague verbs and session-internal jargon, requiring instead that every claim in a message is backed by evidence found in the actual code diff.
Was ein agent damit machen kann
- 1Generate a structured commit messageWrite a git commit message that follows the 17-step analysis and specific formatting rules for my staged changes.
- 2Review existing commit messagesCheck my recent commit messages to ensure they meet the technical specificity and accuracy requirements.
- 3Verify changes against the diffAnalyze my staged changes to ensure the commit message accurately reflects the code and contains no unsupported claims.
Geschrieben basierend auf dem SKILL.md des skills. Jeder use case wurde gegen ein zitat daraus geprüft.
So installierst du den commit Skill
Jeder Befehl kopiert den Skill-Ordner in das Skills-Verzeichnis des Clients. Starte den Client neu oder beginne eine neue Session, um ihn zu laden.Den commit Skill in Plugin installieren
/plugin marketplace add ahundt/autorun /plugin install ar@autorun
Den commit Skill in Claude Code installieren
~/.claude/skills/commit/SKILL.md
$ npx degit ahundt/autorun/plugins/autorun/skills/commit ~/.claude/skills/commitDen commit Skill in Claude Code (project) installieren
.claude/skills/commit/SKILL.md
$ npx degit ahundt/autorun/plugins/autorun/skills/commit .claude/skills/commitDen commit Skill in Codex installieren
~/.agents/skills/commit/SKILL.md
$ npx degit ahundt/autorun/plugins/autorun/skills/commit ~/.agents/skills/commitDen commit Skill in Gemini CLI installieren
~/.gemini/skills/commit/SKILL.md
$ npx degit ahundt/autorun/plugins/autorun/skills/commit ~/.gemini/skills/commitDen commit Skill in Claude.ai installieren
npx degit ahundt/autorun/plugins/autorun/skills/commit commit && zip -r commit.zip commit # claude.ai → Customize → Skills → + → Create skill → Upload a skill → commit.zip
SKILL.md
Auf GitHub ansehenGit Commit Requirements
Third-party repos: Use .git/info/exclude for local-only ignores (e.g., notes/) without modifying .gitignore.
IMPORTANT: Read this entire process (steps 1-17) before starting any git commit work and you must carefully re-analyze the actual for regressions and errors and complete each numbered item and sub-item step-by-step.
Pre-Git Commit Analysis Process
1. Command Execution
git status- see all untracked filesgit diff --staged- see exactly what changes will be committedgit log -5- see recent commit messages for style consistency and for context
2. Complete Change Analysis
Must analyze ALL commits and changes that will be included:
- Not just latest commit - analyze entire scope
- Both staged and unstaged changes
- All commits from branch divergence point (for PRs)
- Previously staged work, not just immediate changes
3. Mandatory Regression Check
Before writing commit message, review git diff --staged to ensure:
- Every claim in commit message has evidence in the actual diff
- No assumptions about previous behavior - only describe what the diff shows changed
- Subject line accurately reflects the files and scope of actual changes
- All described functionality changes match what's visible in the code diff
- If regressions found: record a task with the harness's task tool and fix it before writing the message
Pre-Git Commit Structure & Format
4. Subject Line Format
- Few/grouped files:
<files>: concrete description of what specifically changed- Example:
scoped_allow.py,test_scoped_allow.py: accept day units in scope durations - Example:
test_install_*.py: pin tree publication under two processes(pattern grouping) - Example:
plugins/autorun/commands/*.md: run every bash block through uv run --project(directory grouping)
- Example:
- Many files:
type(scope): concrete description of what specifically changed(scope can be app name)- Example:
fix(installer): retire a hashless legacy tree under --force after a backup
- Example:
- Mixed:
type(scope) <files>: concrete description- Example:
fix(installer) fs.py: republish a hashless legacy tree after backing it up
- Example:
- Always include: Concrete and actionable description of what specifically changed
- Avoid vague terms: "improve performance" → "hydrate ThreadSafeDB once per session instead of once per hook", "enhance security" → "reject relative export paths in plan_export before writing", "update system" → "default the cache gate to off in its FeatureToggle"
5. Message Structure
- Summary first: Concise summary line at top (following format above)
- Previous behavior: Describe what existed before (based on actual git diff, not assumptions)
- What changed: Specific changes made
- Why: Rationale for the changes
- Specific files: List affected files and what changed in each
Pre-Git Commit Content Requirements
6. Concrete & Actionable
- Use specific, measurable descriptions
- Describe functionality that can be tested/verified
- AVOID vague terms: "improved", "enhanced", "HYBRID approach", invented jargon
- USE concrete action words: "fix", "add", "remove", "enable", "disable"
- "update" requires specificity: Use "update X to Y" or "update X by doing Y" - never just "update"
- Include actionable details about what the code now does
- Show exact changes: Before/after comparisons with specific file paths, line numbers
7. Technical Specificity
- Name specific functions, classes, methods affected
- Include file paths and what changed in each file
- Describe technical implementation details
- Mention configuration changes, new dependencies, etc.
8. Accurate Change Classification
- "add" = wholly new feature
- "update X to Y" or "update X by doing Y" = enhancement to existing feature (must specify what changed)
- "fix" = bug fix
- "refactor" = code restructuring
- Must accurately reflect the nature of changes
Pre-Git Commit Context & Documentation
9. Complete Context
- Describe both before/after states
- Explain the problem being solved
- Include enough detail for future developers to understand
- Connect changes to overall system architecture
- Show how changes fit into broader development work
10. Repository Consistency
- Follow existing commit message style from
git log - Match repository's commit message patterns
- Focus on "why" rather than "what" (1-2 sentences for summary)
Pre-Git Commit Security & Quality
11. Security Check
Explicitly check for and prevent committing:
- Secrets or API keys
- Passwords or tokens
- Any sensitive information
12. Testable Outcomes
- Include specific ways to verify changes work
- Mention new functionality that can be tested
- Reference specific commands or use cases enabled
Pre-Git Commit Validation & Quality Control
13. Accuracy Validation Checklist
Before committing, verify:
- Subject format matches
<files>:,type(scope):, ortype(scope) <files>:convention - Every "previous behavior" claim is supported by git diff evidence
- All "what changed" statements match actual lines in git diff --staged
- No claims about functionality that isn't visible in the diff
- Testable outcomes can be verified by running the described commands
- File paths and line numbers are accurate
- Technical details (function names, configurations) match the actual code changes
Pre-Git Development Process Exclusions
14. Avoid Development Methodology References
- Don't mention Claude or AI assistance in development process
- Don't mention multi-agent development methodology used to create the code
- Don't describe HOW the code was developed by assistants
- Don't describe conversational changes or internal commit development process
- Focus on WHAT was built and WHY (multi-agent systems as software features are fine)
15. Avoid Overconfidence & Vague Language
- Never use vague qualifiers: "comprehensive", "complete", "thorough", "HYBRID approach"
- Don't invent terminology: Avoid made-up technical terms that aren't standard
- Avoid abstract descriptions: Use concrete language instead of conceptual descriptions
- Don't bury the main point: Put the key change upfront, not buried in paragraphs
- Avoid absolute claims unless verifiable in the diff
- Don't claim to have "fixed all issues" or "improved everything"
- Use specific, measurable language about actual changes made
- Balance user impact with technical details: Describe both what broke/was fixed AND the implementation approach used
- Acknowledge limitations and scope of changes when relevant
16. Focus on Commit Outcome vs Process
- Describe the whole commit outcome compared to the previous commit state
- Avoid describing incremental conversation steps or iterative development
- Present the final state achieved rather than the journey to get there
- Focus on the complete functional change delivered
Pre-Git Commit Hook Handling
17. Hook Integration
- If pre-commit hooks modify files during commit, retry commit ONCE
- If commit succeeds but hooks modified files, MUST amend commit
- Never use interactive git commands (
-iflag)
Common Git Commit Message Pitfalls & Solutions
Based on analysis of problematic commit messages, avoid these common mistakes:
❌ Bad Commit Message Patterns
-
Vague subject lines: "HYBRID approach", "improve system", "enhance features"
- Problem: Meaningless to someone reading git log
- Solution: Use concrete actions like "fix authentication", "add config validation"
-
Invented jargon: Creating terms like "HYBRID approach" that aren't standard
- Problem: Readers can't understand what was actually implemented
- Solution: Use established technical terms or explain new concepts clearly
-
Missing application context: Subject could apply to any project
- Problem: Lost context about which application was changed
- Solution: Name the component in the scope or lead with its files:
fix(installer): ...,fs.py,traversal.py: ...
-
Buried file impact: Mentioning files deep in message body
- Problem: Hard to understand scope of changes
- Solution: List affected files prominently with brief descriptions
-
Abstract problem descriptions: Conceptual language instead of concrete issues
- Problem: Unclear what the fix actually accomplished
- Solution: Describe specific symptoms and measurable outcomes
✅ Good Git Commit Message Template
type(scope) <files>: [concrete action] [specific component] by [exact method]
Summary: [Concrete action oriented brief high-level description of the change and why it matters]
Previous behavior: [Concrete description of observable behavior and/or limitation(s) being addressed based on git diff]
What changed: [Bulleted list of exact changes with file paths]
- file1.ext: [specific change made]
- file2.ext: [specific change made]
Why: [Root cause explanation in simple terms]
Files affected:
- [list of all modified files with brief description]
Testable: [Specific commands to verify the fix works]
This process ensures git commits are self-documenting, technically precise, security-conscious, and contextually complete while maintaining proper git workflow practices.
Dateien in diesem Skill
1commit Skill-Fragen
Wie installiere ich den commit Skill in Claude Code?
Führe npx degit ahundt/autorun/plugins/autorun/skills/commit ~/.claude/skills/commit aus. Claude Code lädt Skills aus ~/.claude/skills (alle Projekte) oder .claude/skills (ein Projekt) und nutzt einen Skill, wenn seine Beschreibung zur Aufgabe passt.
Welche Clients können den commit Skill nutzen?
Jeder Client, der das Agent Skills-Format liest: Claude Code, Codex, Gemini CLI und claude.ai (als ZIP hochladen). Skripte im Skill laufen nur dort, wo der Client Codeausführung erlaubt.