Optimizes Claude Code skills through iterative testing, binary evaluation, and automated instruction mutation.
Skill Tuner applies the Karpathy autoresearch pattern to systematically evolve and improve your Claude Code skills. By running a target skill against diverse test prompts and evaluating the outputs against strict binary criteria, the tool automatically mutates and refines the SKILL.md instructions to maximize performance. It transforms the prompt engineering process from a manual 'trial-and-error' approach into a data-driven, self-improving optimization loop, complete with performance logging and a live monitoring dashboard.
Key Features
010 GitHub stars
02Optional live dashboard for real-time visualization of score progression
03Intelligent instruction mutation based on highest-scoring historical versions
04Automated evaluation loops using objective binary (Pass/Fail) criteria
05Built-in comparison tools to review diffs between original and optimized skills
06Comprehensive data logging of all runs, scores, and raw outputs
Use Cases
01Increasing the reliability and consistency of complex skill outputs
02Benchmarking skill performance across a wide range of edge cases
03Auto-tuning prompt instructions to meet specific project constraints
