skill-creator

Automate Claude skill creation by drafting a SKILL.md with frontmatter, defining an eval plan, and running an iterative evaluation loop to test trigger accuracy, correctness, and resource usage. This structured workflow helps you rapidly ideate, test, and optimize modular AI capabilities.

An iterative evaluation loop for prompt engineering is a repeatable process that runs prompt-trigger tests, collects performance metrics, and suggests improvements. It allows you to benchmark description tuning and track skill behavior across multiple iterations until performance stabilizes.

Yes, you need PyYAML installed in your environment to support the scripts required by this workflow. The skill relies on these dependencies to automate the creation, refinement, and benchmarking of your modular AI capabilities.

Set up an evaluation framework by defining evaluation criteria alongside your SKILL.md instructions. The framework measures triggering accuracy, correctness, and resource usage, generating baseline and improved descriptions to compare performance and surface actionable insights.

Yes, you can track skill performance over multiple iterations. The workflow maintains a multi-iteration history of prompt-trigger tests and metrics, ensuring generalization and informing future refinements for your AI skills.

The best way to tune skill descriptions is through benchmarking and description tuning. Generate baseline and improved descriptions, compare their performance using the evaluation framework, and apply actionable insights to optimize trigger accuracy.