Overview

A simple, open format for evaluation-driven Claude Code skill development.

Skillet helps you capture failures, run systematic evaluations, and iterate on your skills with data—not hunches.

Why Skillet?

Claude Code skills are powerful, but often don't have the context they need to do real work reliably. Skillet solves this by giving you tools to:

For skill authors: Build capabilities once and test them systematically across multiple scenarios.

For teams: Capture organizational knowledge in portable, version-controlled evaluation suites.

For production: Know your skills work before deploying them, with quantitative feedback on improvements.

What can Skillet enable?

• Capture gaps: Record when Claude Code fails or produces suboptimal results. Turn frustrations into test cases.
• Run evaluations: Systematically test your skills against captured scenarios. Get quantitative feedback on improvements.
• Iterate with data: Make targeted improvements based on actual failure modes.
• Ship with confidence: Know your skills work before deploying them to production.

Quick Start

Install Skillet:

pip install pyskillet

Then capture evals, create a skill, and tune it:

# In Claude Code, capture a failure with /skillet:add
# Then run baseline eval
skillet eval my-skill

# Create a skill from captured evals
skillet create my-skill

# Eval with the skill
skillet eval my-skill ~/.claude/skills/my-skill

# Iteratively improve
skillet tune my-skill ~/.claude/skills/my-skill

Get started

Getting Started

Step-by-step guide to your first evaluation-driven skill.

CLI Reference

Complete documentation for all Skillet commands.

View on GitHub

Browse the source code, report issues, and contribute.