skill-creator

Evaluate Claude skills by running disciplined eval loops with representative prompts, benchmarking multiple versions, and refining descriptions for accurate triggering. Analyze the results to propose improvements to scripts, references, and assets.

Iterate on AI skill descriptions by running structured evaluation loops, reviewing benchmark results, and optimizing the text for better triggering accuracy. This feedback loop ensures reproducible skill improvements across the lifecycle.

Create a new skill by drafting a SKILL.md file from a template, running an evaluation loop with representative prompts, and reviewing the results. Iterate on the implementation and description based on this structured feedback.

Yes, the Skill Creator workflow requires the pyyaml dependency to parse configuration files and manage skill metadata. This allows the evaluation scripts to process the structured definitions during benchmarking.

AI skills fail to trigger correctly when their descriptions lack the necessary keywords or context for accurate matching. Resolve this by running description optimization evals and refining the text based on benchmark performance.

Yes, you can benchmark multiple versions of a skill by running comparative evals with representative prompts. The workflow supports experiment tracking and feedback loops to identify the best performing implementation.