skill-creator

To create and evaluate Claude skills, use a structured workflow to draft a SKILL.md with required frontmatter, generate test prompts, and run trigger and functional evaluations to ensure reliable activation for real user intents.

A SKILL.md evaluation harness generates test prompts and runs parallel with-skill and baseline comparisons, grading outputs to aggregate benchmark reports with timing and token usage for performance measurement.

Optimize AI skill descriptions by iterating on frontmatter text and running train/test split evaluations, which prevents overfitting and ensures the skill triggers appropriately for intended user intents.

Yes, you can use Python with the anthropic client and PyYAML dependencies to automate skill generation, run evaluation loops, aggregate benchmark artifacts, and support description optimization.

The best way to package self-contained assistant skills is to finalize the SKILL.md frontmatter, save test prompts to evals/evals.json, complete the description optimization loop, and aggregate reviewer artifacts.