skill-creator

To create and benchmark Claude skills, draft a SKILL.md, generate evaluation prompts, execute baseline and with-skill runs, then grade outputs to aggregate benchmark statistics for measurable performance differences.

Improving trigger accuracy involves revising the existing skill based on evaluation feedback, then optionally optimizing the description using trigger-eval sets to achieve higher triggering accuracy and repeatable behavior.

Prompt testing works by running prompts with and without the skill, grading the results based on structured expectations and assertions, and aggregating benchmark statistics to evaluate reliable performance comparisons.

Yes, you need Python3 and Claude dependencies to execute the evaluation runs, grade outputs, and perform the iteration loop required for reliable skill benchmarking and performance comparisons.

Use an iteration loop for skill creation when you need repeatable skill behavior across multiple test cases, requiring structured eval setup to repeatedly cycle through drafts and evaluations until the skill is good enough.

Yes, you can benchmark multiple test cases by applying trigger-eval sets and aggregating benchmark statistics, allowing you to quantify performance differences and optionally optimize description triggering accuracy.