skill-creator-cli

To create Claude skills end-to-end, you capture your initial intent, draft the SKILL.md file, run trigger evaluations, and iteratively refine the description and content to maximize triggering accuracy.

The best way to optimize trigger descriptions is to run trigger evaluations, generate improved descriptions based on the results, and track progress across iterations until the skill reliably fires for target prompts.

Trigger evaluation works by testing your drafted SKILL.md against various prompts, allowing you to measure triggering accuracy, identify gaps in the description, and iteratively improve the content for better matching.

To package a Claude skill for deployment, you validate the skill's frontmatter, manage evaluations, add any optional resources, and prepare the final distributable skill bundle for production use.

Yes, you can validate skill frontmatter and manage evals independently, as this skill operates without external dependencies to guide the entire skill lifecycle from drafting to packaging.

A skill description fails to trigger reliably when it lacks optimization, which you can resolve by running iterative trigger evaluations and refining the description until it consistently matches target prompts.