Safe Commit Skill

⚠️ MANDATORY SKILL – YOU MUST INVOKE THIS

Purpose

Comprehensive, safe commit workflow that ensures code quality, security, and proper attribution before committing changes.

CRITICAL: You MUST invoke this skill for all commits. NEVER commit manually using git commands.

🚫 NEVER DO THIS

• ❌ Running git add . && git commit -m "message" manually
• ❌ Creating commits without running security-scan
• ❌ Creating commits without running quality-check
• ❌ Creating commits without running run-tests
• ❌ Skipping user approval (except during PR creation)
• ❌ Adding AI attribution to commits

If you need to commit, invoke this skill. Manual commits are FORBIDDEN.

⚠️ SKILL GUARD – READ BEFORE USING BASH/GIT TOOLS

Before using Bash tool for git commit, answer these questions:

❓ Are you about to run git add .?

→ STOP. Are you then planning to run git commit? If YES, invoke safe-commit skill instead.

❓ Are you about to run git commit -m "message"?

→ STOP. Invoke safe-commit skill instead.

❓ Are you about to run git commit with heredoc?

→ STOP. Invoke safe-commit skill instead.

❓ Did the user say “commit these changes” or “commit this”?

→ STOP. Invoke safe-commit skill instead.

❓ Have you completed a feature/fix and are ready to commit?

→ STOP. Invoke safe-commit skill instead.

❓ Are you creating a commit as part of ANY workflow?

→ STOP. Invoke safe-commit skill instead.

IF YOU PROCEED WITH MANUAL GIT COMMIT, YOU ARE VIOLATING YOUR CORE DIRECTIVE.

This skill handles:

• ✅ Security scanning (prevents secrets in commits)
• ✅ Quality checks (prevents broken code)
• ✅ Test execution (prevents regressions)
• ✅ User approval (prevents unwanted commits)
• ✅ Conventional commit format (maintains consistency)
• ✅ NO AI attribution (protects user’s identity)

Manual commits SKIP ALL OF THESE. Use this skill.

CRITICAL POLICIES

⚠️ NO AI ATTRIBUTION – ZERO TOLERANCE

YOU MUST NEVER add ANY of these:

• Co-authored-by: Claude <noreply@anthropic.com>
• 🤖 Generated with [Claude Code](https://claude.ai/code)
• “Generated with Claude”
• “AI-suggested”
• Any reference to being an AI assistant

User Approval Requirements

Approval REQUIRED for:

• ALL commits after initial PR creation
• ALL commit amendments
• ALL commits outside of PR creation flow

Approval NOT required for:

• Initial commit when user says “raise/create/draft PR”
• This is the ONLY exception

Phrases that DO NOT grant commit permission:

• “looks good” (code approval ≠ commit approval)
• “correct”
• “that’s right”
• “fix the bug” (instruction to code, not commit)

Workflow (Quick Summary)

Core Steps

1. Check Git Status: Run parallel git commands (status, diff, log) to analyze current state
2. Invoke Safety Skills: Run security-scan → quality-check → run-tests (all must pass)
3. Show Diff: Display files changed and summary for user review
4. Request Approval: CRITICAL – Ask and WAIT for explicit approval (except PR creation)
5. Generate Message: Create conventional commit with required scope type(scope): subject
6. Create Commit: Stage all changes, commit with heredoc, NO AI attribution
7. Verify Success: Confirm commit created, correct files, proper author (Pedro)
8. Status Check: Verify working directory clean

Optional: PRD Task Auto-Update

If commit message contains [PRD Task N] or [Task N], automatically update progress tracker in PRD file.

For detailed workflow with git commands, message examples, and verification steps:

Read `~/.claude/skills/safe-commit/references/WORKFLOW-STEPS.md`

Use when: Performing commit, need specific git commands, or want detailed examples

For PRD task auto-update details:

Read `~/.claude/skills/safe-commit/references/PRD-TASK-UPDATE.md`

Use when: Working with PRD tracking or implementing progress automation

For pre-commit hook handling:

Read `~/.claude/skills/safe-commit/references/PRE-COMMIT-HOOKS.md`

Use when: Dealing with hook-modified files or commit amendment scenarios

Integration with Other Skills

This skill invokes:

• security-scan – Step 2.1
• quality-check – Step 2.2
• run-tests – Step 2.3

This skill is invoked by:

• create-pr – As part of PR creation workflow

Exception: PR Creation Flow

When invoked by create-pr skill:

• Skip Step 4 (user approval)
• Proceed directly to commit
• This is the ONLY time auto-commit is allowed

The create-pr skill is only invoked when user explicitly says “raise/create/draft PR”

Error Handling

If security scan fails:

❌ Cannot commit: Security issues detected

[Details from security-scan skill]

Please fix security issues and try again.

If quality check fails:

❌ Cannot commit: Code quality issues detected

[Details from quality-check skill]

Please fix linter/formatter issues and try again.

If tests fail:

❌ Cannot commit: Tests failing or coverage below threshold

[Details from run-tests skill]

Please fix failing tests and improve coverage, then try again.

If git commit fails:

❌ Commit failed

Error: [git error message]

Possible causes:
- Pre-commit hook failure
- Git configuration issue
- File system permissions

Please investigate and retry.

Best Practices

1. Always run in order – Security → Quality → Tests → Commit
2. No skipping checks – All must pass
3. Get explicit approval – Don’t assume permission (except PR creation)
4. Descriptive messages – Help future you understand why
5. Proper scopes – Never omit scope from commit message
6. Verify attribution – Always ensure Pedro is sole author
7. Clean commits – Stage all changes, commit once

Commit Message Quality Checklist

Before committing, verify message has:

• ✅ Type and scope in format: type(scope):
• ✅ Imperative mood in subject
• ✅ Subject ≤ 50 characters
• ✅ Body explains why (if needed)
• ✅ References issues/tickets (if applicable)
• ✅ NO AI attribution anywhere
• ✅ NO Co-authored-by tags

Emergency Override

If user explicitly states “force commit” or “skip checks”:

YOU MUST:

1. Warn about risks
2. List which checks are being skipped
3. Get explicit re-confirmation
4. Document in commit message what was skipped
5. Create follow-up ticket for remediation

This should be EXTREMELY RARE.