skill-creator

To create and optimize Claude skills, draft a SKILL.md file, assemble supporting scripts and references, run trigger evaluations, and iterate on descriptions based on collected metrics to ensure reliable activation.

A trigger evaluation loop tests skill activation by running prompts, capturing performance metrics, and refining the triggering description to maximize accurate activation while avoiding overfitting.

Yes, PyYAML is required as a dependency to parse and structure your skill artifacts, allowing you to assemble scripts, references, and assets into a repeatable workflow for future iterations.

The best way to organize assets is to scaffold them alongside your SKILL.md and scripts, grouping code, documentation, and references into a structured workflow that supports detailed feedback generation.

Your Claude skill fails to trigger reliably when the description is underfitting or overfitting; you must run trigger-eval loops, review metrics, and iterate on the description to improve activation accuracy.

Yes, skill-creator is designed to generate detailed, structured feedback during the evaluation and iteration workflow, producing output optimized for vector embeddings across multiple skills.