darwin-skill

What problem does it solve?

Darwin.skill provides an automated framework for evaluating and evolving Claude-style SKILL.md entries. It orchestrates a rigorous, repeatable optimization loop that examines both the static structure of SKILL.md and the real-world effectiveness of its guidance, ensuring skills actually improve over time rather than just looking better on paper.

Core Features & Use Cases

• Dual evaluation using an 8-dimension rubric (structure 60 points, effectiveness 40 points) plus live testing with designed test prompts.
• Git-backed hill-climbing optimization that only keeps improvements through commits and reverts.
• Human-in-the-loop, pausing after each optimization cycle with diff and results card generation for review.
• Test-prompt design per skill and automated generation of visual result cards to communicate progress.
• Central results log (results.tsv) for auditing, baseline tracking, and future resumption.

Quick Start

Install darwin.skill and start the optimization workflow by telling your agent to optimize all skills.