Sensei
AI-powered skill frontmatter compliance improver for GitHub Copilot CLI and Claude Code 🥋
Sensei automates the improvement of Agent Skills frontmatter compliance using the Ralph loop pattern — iteratively refining skills until they reach Medium-High compliance with all tests passing. Works with both GitHub Copilot CLI and Claude Code.
It solves the “skill collision” problem where AI agents invoke the wrong skill for a given prompt because of missing triggers, vague descriptions, or token bloat. Sensei reads your skill, scores its compliance, improves the frontmatter with proper trigger phrases, runs tests, checks token budgets, and repeats until the skill is properly optimized.
Built it because managing dozens of agent skills manually was painful — Sensei does the tedious compliance work so you can focus on building delightful skills. It even includes GEPA evolutionary optimization for deep automated improvements. 🧬
Level up your agent skills with an AI-powered sensei that iteratively polishes your skill frontmatter until it’s flawless and ready for action! Say goodbye to skill collisions and token bloat with smart, automated improvements that keep your skills sharp and context-friendly.
Key features:
- 🥋 Iterative refinement loop for frontmatter compliance
- 🚀 Fast and deep optimization modes including GEPA evolutionary improvements
- 🧪 Automated testing and scoring to ensure quality
- 📊 Token usage analysis with actionable suggestions
This summary was generated by GitHub Copilot based on the project README.