Git Skill

Code is Mechanism, Prompt is Policy

Architecture

This skill uses @skill_command decorator in scripts/*.py files. Commands are automatically exposed via MCP as git.command_name.

Available Commands

Smart Commit Workflow

Use git.smart_commit for secure, human-in-the-loop commits:

# Step 1: Start workflow
git.smart_commit(action="start")
# Returns workflow_id and diff preview

# Step 2: After LLM analysis and user approval
git.smart_commit(action="approve", workflow_id="xxx", message="feat: description")

Flow: stage_and_scan → route_prepare → format_review → re_stage → interrupt → commit

Staged Files Feature

Stage and Scan Workflow

The stage_and_scan function provides automatic staging with security validation:

Stage All Files → Security Scan → Lefthook Pre-commit → Finalize

Key Features

1. Automatic Staging

   stage_and_scan(project_root=".")
   # Returns: {staged_files, diff, security_issues, lefthook_error}
   

2. Security Scanning

   • Detects sensitive files (.env*, *.pem, *.key, *.secret, etc.)
   • Automatically un-stages detected files
   • Returns list of security issues

3. Lefthook Integration

   • Runs pre-commit hooks after staging
   • Re-stages files modified by lefthook formatters
   • Returns lefthook output for review

Staged Files Commands

Security Patterns Detected

.env*, *.env*, *.pem, *.key, *.secret, *.credentials*
id_rsa*, id_ed25519*, *.priv
secrets.yml, secrets.yaml, credentials.yml

Usage Guidelines

Read Operations (Safe - Use Claude-native bash)

git status
git diff --cached
git diff
git log --oneline

Write Operations (Use MCP Tools)

Key Principle

Read = Claude-native bash. Write = MCP tools.