skills/skill-creator

🔗 View on GitHub: anthropics/skills/skill-creator

🚀 Quick Installation

Create new skills, modify and improve existing skills, and measure skill performance.

Claude Code Plugin

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

What It Does

A comprehensive skill for creating, improving, and evaluating Agent Skills. Use this when you want to:

• ✨ Create a skill from scratch
• 🔧 Edit or optimize an existing skill
• 📊 Run evals to test a skill
• 📈 Benchmark skill performance with variance analysis
• 🎯 Optimize a skill's description for better triggering accuracy

The Skill Creation Process

1. Decide — What should the skill do and how should it do it?
2. Draft — Write the initial SKILL.md
3. Test — Create test prompts and run Claude with the skill
4. Evaluate — Review results qualitatively and quantitatively
5. Iterate — Rewrite based on feedback
6. Scale — Expand the test set and try again at larger scale

Key Features

Skill Structure

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (name, description required)
│   └── Markdown instructions
└── Bundled Resources (optional)
    ├── scripts/     - Executable code for deterministic tasks
    ├── references/  - Docs loaded into context as needed
    └── assets/      - Templates, icons, fonts

Evaluation System

• Parallel Testing — Run with-skill vs baseline simultaneously
• Quantitative Metrics — Token usage, timing, pass rates
• Qualitative Review — Browser-based eval viewer for human feedback
• Benchmark Analysis — Statistical comparison with variance analysis

Description Optimization

• Generate 20 realistic eval queries (should-trigger vs should-not-trigger)
• Run automated optimization loop (up to 5 iterations)
• Select best description by test score (not train score)

Best Practices

• ✅ Keep SKILL.md under 500 lines
• ✅ Use imperative form in instructions
• ✅ Explain the "why" behind instructions
• ✅ Include realistic examples
• ✅ Make descriptions "pushy" to combat undertriggering
• ❌ Avoid heavy-handed MUSTs and rigid structures

Example Workflow

"I want to make a skill for X"
↓
Interview → Draft → Test → Evaluate → Iterate → Package

📚 Creating Custom Skills | 🎓 Agent Skills Spec