git-claw

English · 한국어

An Agent Skill that enforces consistent Git commit messages, PR templates, and AI review analysis across projects.

Works with Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot, Amp, and 40+ agents.

Installation

Option 1: Claude Code Plugin (Recommended)

# Add marketplace
/plugin marketplace add chanmuzi/git-conventions

# Install plugin
/plugin install git-conventions@git-conventions

Option 2: Skills CLI (All Agents)

npx skills add chanmuzi/git-conventions

Included skills:

The interactive installer lets you select skills, target agents, scope (project/global), and install method.

To install all skills at once without prompts:

npx skills add chanmuzi/git-conventions --skill commit --skill pr --skill issue --skill review-reply --skill code-review --skill handoff -g

Update

Recommended: Enable auto-update to receive new features and fixes automatically.
See platform-specific instructions below.

Claude Code Plugin

# Update the plugin
/plugin update git-conventions@git-conventions

Or enable auto-update for hands-free upgrades (recommended):

/plugin → Marketplaces tab → select marketplace → Enable auto-update

Skills CLI

# Check for available updates
npx skills check

# Update all installed skills
npx skills update

If you installed with Symlink (recommended), a single update applies to all agents at once. Copy installs require updating each copy individually.

Skills CLI does not auto-update. Run npx skills check periodically, or after major releases announced in the changelog.

Skills

Skill prefix by agent: Claude Code uses / (e.g., /commit), Codex CLI uses $ (e.g., $commit).

/commit — Create a Git Commit

Analyzes your staged/unstaged changes and proposes a commit message following the conventional commit format. Automatically detects when the current changes overlap with the last commit and amends instead of creating a new commit.

/commit              # Analyze and commit (with amend auto-detection)
/commit --no-amend   # Skip amend detection; always create a new commit

Commit message format:

{type}: {description}
{type}({scope}): {description}

Types: feat, fix, refactor, style, docs, test, perf, chore, hotfix

/pr — Create a Pull Request

Creates a PR with a structured template. Automatically detects whether to use the Individual PR template or the Release PR template.

/pr              # Individual PR (Feat, Fix, Refactor, etc.)
/pr -g           # Include Mermaid change-flow graph
/pr release      # Release PR (dev → main integration)

PR title format:

{Type}: {description}               # Individual (e.g., Feat: add new feature)
Release: dev → main 통합 (vX.Y.Z)  # Release

/issue — Create a GitHub Issue

Creates a structured issue with the appropriate template and auto-assigns type/priority labels.

/issue             # General issue (type inferred from context)
/issue bug         # Bug report
/issue feature     # Feature request

Templates: Bug Report, Feature Request, General
Auto-labeling: bug, feature, enhancement, critical/high/medium/low, etc.

/review-reply — Review & Reply to PR Comments

Collects review comments (CodeRabbit, Copilot, teammates, etc.) from a PR, analyzes their validity against the actual code, and discusses findings with you.

/review-reply          # Review current branch's PR
/review-reply 42       # Review PR #42

/code-review — Context-Aware Code Review

Analyzes PR or local code changes using a multi-agent pipeline with domain-specific agents (Security, Performance, Architecture, Domain Logic). Cross-validates findings against code context to filter false positives, then produces severity-based structured output.

/code-review           # Auto-detect PR on current branch, or review working dir
/code-review 42        # Review PR #42
/code-review src/auth/ # Review specific path
/code-review --wd      # Force working dir review (skip PR auto-detection)

Flags:

• --wd — Force working directory mode, even on a PR branch
• --domain security,perf — Override auto-detected domains
• -y / -f — Publish without approval
• -g — Generate Mermaid change-flow graph (PR mode only)
• -q / --quick — Quick mode: single-pass analysis (no agent spawn), max 2 domains, Critical/Warning first (Info fallback when none found)
• --no-codex — Disable Codex integration
• --codex — Run both Codex review and adversarial review
• --codex-general — Use Codex general review only (without adversarial)

Natural language: You can use flags or natural language — Claude translates your intent into the appropriate flags before invoking the skill.

/code-review 42 바로 올려줘             # → /code-review 42 -y
/code-review security만 봐줘            # → /code-review --domain security
/code-review codex 둘 다 돌려줘         # → /code-review --codex
/code-review codex 없이 리뷰해줘        # → /code-review --no-codex
/code-review working dir로 봐줘         # → /code-review --wd
/code-review 간단하게 봐줘              # → /code-review --quick

Codex integration (optional, recommended): When the Codex plugin is installed and authenticated, /code-review automatically runs Codex adversarial review in parallel with domain agents. Findings are cross-validated and merged with source tags. Use --no-codex to opt out.

To set up Codex integration:

1. Install the Codex plugin: claude plugin add codex
2. Run !codex setup in Claude Code to authenticate with your OpenAI API key
3. /code-review will automatically detect and use Codex — no additional flags needed

Inline review comments: In PR mode, findings are published as inline review comments attached to specific diff lines via the GitHub Review API. Each finding becomes an independent conversation thread — reply, resolve, or apply suggestions individually. A severity summary table is included in the review body. Findings outside the diff are included as "General Findings" in the summary.

Severity levels: 🔴 Critical, 🟡 Warning, 🟢 Info

/handoff — Session Handoff Prompt

Generates a copy-ready handoff prompt that transfers work context to a new session. Auto-detects artifacts, git state, and conversation context, then structures a concise prompt following the reference-not-repeat principle.

/handoff                   # Auto-detect and generate handoff prompt
/handoff -y                # Skip confirmation, output directly
/handoff auth refactoring  # Focus handoff on a specific topic

Detection cascade: Artifacts (.omc/specs/, .omc/plans/) → Git state → Conversation context

Skill recommendation: When plugins like OMC or Codex are detected, the handoff recommends the most appropriate skill for the next session (e.g., /autopilot, /ralph, /commit).

Output: Terminal text optimized for /copy. Never repeats artifact content — references file paths instead.

Language Behavior

All commands write output (commit messages, PR titles/body) in the language configured in your project's CLAUDE.md. If no language is set, the user's conversational language is used.

Technical terms are kept in their original form to preserve clarity — no forced translations.

Label System

/pr and /issue share a unified label scheme. Labels are auto-created on first use — if a label already exists, it is left unchanged.

• No namespace prefix — clean, simple names (bug instead of type: bug)
• GitHub-standard colors — matches GitHub's default palette where applicable
• Conflict-safe — gh label create ... 2>/dev/null || true (create if missing, skip if exists)

Conventions at a Glance

Integration with CLAUDE.md

Add the following to your global ~/.claude/CLAUDE.md to reference these conventions:

## Git Conventions
- Commit: `{type}: {description}` (lowercase prefix: feat, fix, refactor, style, docs, test, perf, chore, hotfix)
- Branch: `{type}/{english-kebab-case}` (feat/, fix/, refactor/, docs/, hotfix/)
- PR title: `{Type}: {description}` (capitalized prefix: Feat, Fix, Refactor, Perf, etc.)
- Release PR: `Release: dev → main 통합 (vX.Y.Z)`
- Use `/commit`, `/pr`, `/pr release`, `/issue`, `/review-reply`, `/code-review`, `/handoff` commands for full workflows

License

MIT