skill-creator

To create and test AI skills, you can automate drafting SKILL.md files, generate test prompts, and run with-skill versus baseline evaluations to capture timing, tokens, and aggregate benchmark metrics.

The workflow for iteratively improving AI skill descriptions involves running comparative evals, capturing reviewer feedback, and applying guided prompts to rewrite skills and optimize triggering accuracy.

Yes, you can run comparative evaluations for prompt engineering locally by using runnable Python utilities to orchestrate baseline runs, capture timing and token data, and output grader-compatible JSON.

Test automation for skill creation relies on local Python utilities for scripts and viewers, operating without external dependencies to package hooks and aggregate benchmark metrics across iterations.

To package AI skills for distribution after benchmarking, you use the iteration workflow to apply reviewer feedback, optimize descriptions for triggering accuracy, and finalize the skill metadata.

The best way to evaluate regressions in AI skill metadata is to run parallel with-skill and baseline tests, grade assertions, and surface variance and regressions through aggregated benchmark metrics.