git-commit-message（git 提交消息）

Git Commit Message Generator

Auto-generates conventional commit messages from git diffs with tiered format enforcement

Purpose

Analyze staged git changes and generate concise, meaningful commit messages following a tiered Conventional Commits specification. This skill examines file modifications, additions, and deletions to infer the type and scope of changes, producing commit messages that match the importance of the change - from detailed documentation for critical features to concise messages for minor updates.

Key Innovation: Three-tier format system that balances thoroughness for critical commits (feat, fix, security) with efficiency for routine changes (docs, chore, style).

When This Skill Activates

• When /commit-msg command is invoked
• When invoked from a commit-msg/prepare-commit-msg hook (if installed)
• When user requests commit message suggestions
• When analyzing changes before creating a commit

Core Capabilities

1. Diff Analysis

• Parse git diff --staged output
• Identify modified, added, and deleted files
• Analyze code changes (additions, deletions, modifications)
• Detect patterns across multiple files

2. Change Classification

• Determine commit type from changes:

- feat: New features or functionality - fix: Bug fixes - security: Security fixes or hardening - refactor: Code restructuring without behavior change - docs: Documentation changes - style: Formatting, whitespace, code style - test: Adding or modifying tests - chore: Build process, dependencies, tooling - perf: Performance improvements - ci: CI/CD configuration changes - build: Build system changes - revert: Reverting previous commits

3. Scope Detection

• Infer scope from file paths and patterns:

- Directory names (e.g., api, auth, ui) - File name patterns (e.g., *.test.js → tests) - Framework conventions (e.g., components/, services/)

4. Message Generation

• Format: type(scope): description
• Enforce tier limits: Tier 1 summary max 50 chars; Tier 2/3 summary max 72 chars (ideal 50)
• Use imperative mood ("add" not "added")
• Focus on "what" and "why", not "how"
• Provide 2-3 alternative suggestions

Tier System: Smart Format Enforcement

This skill uses a three-tier format system that matches message detail to commit criticality:

Tier 1: Critical Commits (feat, fix, perf, security)

Requirements: Detailed documentation with impact statement

Format:

type(scope): summary line (max 50 chars)

- Detailed description point 1
- Detailed description point 2
- Detailed description point 3

This change [impact statement describing user-facing benefit or risk addressed].

Affected files/components:
- path/to/file1
- path/to/file2

Why: Features, fixes, and performance changes affect users directly and need thorough documentation for future reference and changelog generation.

Tier 2: Standard Commits (refactor, test, build, ci)

Requirements: Brief context and file list

Format:

type(scope): summary line (max 72 chars)

Brief explanation of what changed and why (1-2 sentences).

Files: path/to/file1, path/to/file2

Why: Internal improvements need context for maintainability but don't require extensive documentation.

Tier 3: Minor Commits (docs, style, chore)

Requirements: Summary line, optional description

Format:

type(scope): summary line (max 72 chars)

[Optional: Additional context if helpful]

Why: Documentation and routine maintenance are self-explanatory from the diff; verbose messages add noise.

Workflow

0. Pre-staging typecheck (if project uses TypeScript):
   - Run `tsc --noEmit` on changed files before staging
   - Fix type errors before committing (avoids pre-commit hook retry loops)
1. Get staged changes (staged only, not working tree):
   - git diff --staged --name-status
   - git diff --staged --stat
   - git diff --staged
2. Load config → frameworks/shared-skills/skills/git-commit-message/config.yaml
3. Analyze changes:
   - Count files modified/added/deleted
   - Identify primary change type using analysis patterns
   - Detect scope from project structure (config.yaml)
   - Determine tier (1/2/3) based on commit type
   - Extract key modifications
4. Generate commit messages:
   - Apply tier-appropriate format
   - Primary suggestion (best match)
   - Alternative 1 (different scope/angle)
   - Alternative 2 (broader/narrower focus)
5. Validate against rules:
   - Check forbidden patterns
   - Verify required elements present
   - Ensure length limits
6. Present to user with explanation and tier info

Optional Modes (If Supported By The Caller)

• --validate "<message>": Validate a commit message without generating suggestions (format/type/scope/length/forbidden patterns; then report required Tier 1/2/3 elements if missing).
• --tier <1|2|3>: Force the tier format (overrides auto-detection).
• --interactive or -i: Ask for confirmation of type, scope, and summary before final output.

Output Format

[NOTE] Suggested Commit Messages (based on X files changed)

PRIMARY:
feat(api): add user authentication endpoints

ALTERNATIVES:
1. feat(auth): implement JWT token validation
2. feat: add user authentication system

ANALYSIS:
- 3 files modified in src/api/
- New functions: authenticateUser, generateToken
- Primary change: new feature (authentication)
- Scope detected: api/auth

Conventional Commits Quick Reference

Type Guidelines:

• feat: User-facing features or API additions
• fix: Corrects incorrect behavior
• refactor: Improves code without changing behavior
• docs: README, comments, documentation files
• style: Formatting only (prettier, eslint --fix)
• test: Test files or test utilities
• chore: Build scripts, package updates, config
• perf: Measurable performance improvements
• ci: GitHub Actions, CircleCI, build pipelines

Scope Guidelines:

• Use lowercase
• Be specific but not too narrow
• Match your project's module structure
• Omit if changes span multiple unrelated areas

Description Guidelines:

• Start with lowercase verb
• No period at the end
• Be specific and concise
• Focus on user impact for feat and fix

Edge Cases

Multiple unrelated changes:

• Suggest splitting into separate commits
• If forced to combine, use broader scope or omit scope

Breaking changes:

• Append exclamation mark after type/scope (example: feat(api)!: change auth flow)
• Include BREAKING CHANGE in body (handled by user)

WIP or experimental:

• Use chore(wip): description or feat(experimental): description

No meaningful changes:

• Detect and warn: "No staged changes detected"
• Suggest git add commands

Integration Points

Pre-commit hook: Triggered before commit (if installed/configured) Slash command: Manual invocation via /commit-msg Direct skill call: From other skills or tools

Best Practices

1. Analyze context: Look at file paths, function names, import statements
2. Prioritize clarity: Prefer obvious descriptions over clever ones
3. Respect conventions: Follow project's existing commit patterns if detected
4. Avoid hallucination: Only describe what's actually in the diff
5. Be concise: 50 chars is ideal, 72 is maximum for first line
6. Stage specific files: Use git add <file1> <file2>, not git add -A or git add., to avoid pulling in unrelated changes or sensitive files
7. Avoid heredoc in sandboxed shells: Sandboxed environments may block temp file creation for here-documents. Use git commit -m "$(cat <<'EOF'\nmessage\nEOF\n)" or pass -m "message" directly
8. Pre-commit typecheck: Run tsc --noEmit on the staged surface before committing to catch type errors early and avoid retry cascades from pre-commit hooks

Example Analyses

Scenario 1: New React component

Files: src/components/UserProfile.tsx, src/components/UserProfile.test.tsx
Changes: +120 lines, component definition, props interface, tests
Message: feat(components): add UserProfile component

Scenario 2: Bug fix in API

Files: src/api/auth.ts
Changes: -5 +8 lines, fix token expiration check
Message: fix(auth): correct token expiration validation

Scenario 3: Documentation update

Files: README.md, docs/api.md
Changes: +45 lines documentation
Message: docs: update API documentation and README

Scenario 4: Dependency update

Files: package.json, package-lock.json
Changes: version bumps for eslint, typescript
Message: chore(deps): update eslint and typescript

Analysis Patterns: Smart Type Detection

The skill uses pattern matching to intelligently detect commit types from diffs:

feat Detection

• New files created (especially in src/, components/, api/)
• New functions/classes exported (export function, export class)
• New API routes (app.get, router.post, etc.)
• New assets/skills (in.claude/, custom-gpt/, etc.)
• Threshold: 20+ lines added typically indicates feature

fix Detection

• Test file changes (often indicates bug reproduction)
• New conditionals (validation fixes)
• Error handling additions (try, catch, throw)
• Input validation (validate, sanitize, check)
• Commit message hints: Words like "bug", "issue", "error", "crash"

refactor Detection

• Balanced changes (similar additions and deletions)
• Function renames/moves (same logic, different location)
• No new features or fixes
• Test coverage unchanged
• Keywords: "extract", "move", "rename", "reorganize"

docs Detection

• File patterns: .md, .txt, README, CHANGELOG, /docs/
• Pure documentation changes (no code modifications)
• Mixed code+docs: Prefer code type, note docs in description

test Detection

• File patterns: test.js, spec.ts, __tests__/, /tests/
• Test framework patterns: describe, it, test, expect, assert

style Detection

• CSS/styling files: .css, .scss, .sass, .less
• Formatter configs: prettier, eslint
• Whitespace-only changes
• Keywords: "formatting", "indent", "whitespace"

chore Detection

• Dependency files: package.json, requirements.txt, Gemfile
• Lock files: package-lock.json, yarn.lock
• Config files: .gitignore, .env
• Keywords: "dependency", "deps", "upgrade", "bump"

Configuration

Project-specific configuration loaded from config.yaml:

• Scope mapping: Maps directory patterns to scope names (e.g., frameworks/claude-code-kit/** → claude-kit)
• Tier rules: Defines which commit types require which tier format
• Forbidden patterns: Blocks commits with generic messages or assistant/tool attribution
• Analysis patterns: Customizes type detection logic for your codebase
• Validation mode: strict (block), warning (warn), or disabled

Forbidden Patterns (Validation)

The skill automatically blocks commits with these patterns:

Generic/Vague Messages

• [FAIL] "Update files" → [OK] "docs: update API reference"
• [FAIL] "Fix stuff" → [OK] "fix(auth): correct token validation"
• [FAIL] "Change code" → [OK] "refactor(utils): simplify date formatting"

Assistant/Tool Attribution (Per Repository Policy)

• [FAIL] "Generated with Claude Code"
• [FAIL] "Co-Authored-By: Claude noreply@anthropic.com"
• [FAIL] Any assistant/tool attribution in commit messages

Work-in-Progress Markers

• [WARNING] "WIP: feature" (warning - should be squashed before merge)
• [WARNING] "temp: quick fix" (warning - should be squashed)

Missing Type

• [FAIL] Commits without type prefix (feat, fix, docs, etc.)

Error Handling

• No staged changes: Run git status and guide user to git add files
• Binary files only: Note that commit message should mention file types
• Merge conflicts: Detect and suggest chore: resolve merge conflicts
• Git not available: Graceful failure with helpful error message
• Forbidden pattern detected: Show error with examples and block commit (strict mode)
• Missing required elements: List what's missing based on tier requirements
• Length exceeded: Show character count and suggest shortening

Integration with Repository

This skill integrates with the AI-Agents repository standards:

• CLAUDE.md reference: Mandatory skill usage before commits
• config.yaml: Project-specific scope mappings and rules
• Pre-commit hook: Automatic activation before git commits
• CONTRIBUTING.md: Commit guidelines for contributors

Commit Message Template

assets/template-commit-message.md — Copy-paste template and good/bad examples.

Use it to standardize type(scope): summary messages and keep history automation-friendly.

Security-Sensitive Commits

assets/template-security-commits.md — Guide for handling security-sensitive changes.

Key Sections

• Pre-Commit Security Checklist — Secrets detection, prohibited patterns
• Security-Related Commit Types — Security fix, enhancement, configuration
• Accidental Secret Commits — Immediate response, rotation, history cleanup
• Sensitive File Patterns —.gitignore templates, files that should never be committed
• Audit Trail Requirements — CVE, CVSS, CWE metadata for security commits

Do / Avoid

GOOD: Do

• Run secrets scan before every commit
• Rotate secrets immediately if exposed
• Use environment variables for credentials
• Document security fixes with CVE/CVSS
• Require security team review for auth changes
• Keep.gitignore updated for secret patterns

BAD: Avoid

• Committing secrets "temporarily"
• Using hardcoded credentials in tests
• Storing real credentials in example files
• Assuming deleted secrets are safe
• Committing before secrets scan completes
• Using generic commit messages for security fixes

Anti-Patterns

Optional: AI/Automation

Note: AI suggestions should preserve human intent.

• Commit message suggestions — Draft from diff analysis
• Type detection — Pattern-based commit type inference
• Scope detection — Auto-detect from changed paths

Bounded Claims

• AI-generated messages need human review and modification
• Automated type detection may miss context
• Security commits always need human judgment

Resources

Version: 2.1.1 Last Updated: 2026-01-26 Repository: AI-Agents (documentation repository) Conventional Commits Spec: https://www.conventionalcommits.org/