skill-creator

To create and refine Claude skills, you draft a skill from a high-level idea, run evaluation prompts to test triggering behavior, analyze the results, and iteratively rewrite the skill to maximize accuracy and usefulness across real-world prompts.

Benchmarking Claude skills involves running evaluation loops that test trigger behavior and collect quantitative metrics, then comparing those results against baselines to quantify improvements and inform better skill descriptions.

Running evaluation prompts helps with prompt engineering by collecting quantitative metrics on how well the Claude skill triggers, allowing you to analyze results, generate reviews, and iteratively rewrite the skill for better performance.

Yes, you need the pyyaml dependency installed to use this skill creator, which guides you through drafting, evaluating, and optimizing Claude workflow skills end-to-end.

Yes, you can draft new Claude skills from a high-level idea and iterate on the draft using structured prompts provided by the skill creator to maximize triggering accuracy and effectiveness.

A Claude skill may not trigger accurately if its description is poorly optimized; you can fix this by running evaluation prompts, analyzing the benchmark results, and iteratively rewriting the skill to improve triggering accuracy.