skill-creator

To create and iterate on AI agent skills, you draft a SKILL.md file and run an interactive evaluation loop that tests prompts, tracks performance, and refines triggering accuracy. This automated process drives systematic improvement through benchmark insights and description optimization.

The best way to optimize skill triggering is by running automated evaluation loops that generate benchmark insights. These eval outcomes allow you to apply focused improvements to the skill's description text, systematically increasing triggering accuracy and reducing under or over-triggering.

You package AI agent skills into a .skill bundle through the Skill's built-in automation, which includes safety checks and validation. This encapsulates your drafted and refined SKILL.md content along with optional tooling like evals and benchmark data into a distributable format.

You can use this Skill for both building new AI agent skills and improving existing ones. It applies an iterative refinement process using structured evals, benchmarks, and description optimization across multi-step workflows to enhance performance regardless of the skill's current state.

Yes, PyYAML is required as a dependency to run the skill evaluation loops and benchmark generation. It supports the underlying automation that orchestrates evals, tracks performance metrics, and handles the SKILL.md content processing.

Your AI agent skill under-triggers when its description lacks optimization based on actual performance data. Running automated evaluation loops generates benchmark insights that identify triggering issues, enabling focused description improvements to increase accuracy.