Skip to content
Back to skills

Safe Commit Majiayu000

ASecurity

⚠️ MANDATORY - YOU MUST invoke this skill when committing. Complete commit workflow with all safety checks. Invokes security-scan, quality-check, and run-tests skills. Shows diff, gets user approval, creates commit with conventional format. NO AI attribution. User approval REQUIRED except during PR creation. NEVER commit manually.

  • 57 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
toolsgobashgitsecurity

Works with

  • claude code

Security analysis

A100/100

Scanned October 17, 2026

npx -y skills add diegosouzapw/awesome-omni-skill --skill safe-commit-majiayu000 --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Safe Commit Majiayu000?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Safe Commit Majiayu000
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/diegosouzapw-safe-commit-majiayu000/badge)](https://www.skillsdirectory.com/skills/diegosouzapw-safe-commit-majiayu000)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: safe-commit
description: ⚠️ MANDATORY - YOU MUST invoke this skill when committing. Complete commit workflow with all safety checks. Invokes security-scan, quality-check, and run-tests skills. Shows diff, gets user approval, creates commit with conventional format. NO AI attribution. User approval REQUIRED except during PR creation. NEVER commit manually.
version: 1.0.1
---

# 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.**

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.