skill-creator

To create and optimize Claude Code skills, you draft SKILL.md content with YAML frontmatter, run trigger-evaluation benchmarks against test prompts, and iteratively refine descriptions based on pass or fail performance.

Trigger optimization for skill authoring reduces false triggers and under-triggering by analyzing benchmark results across should-trigger and should-not-trigger query sets, which is needed when a skill fails to activate reliably.

You benchmark Claude Code skills by generating quantitative eval sets, running test prompts across variants, and analyzing pass or fail patterns to guide description optimization and improve triggering accuracy.

Yes, you can use pyyaml to manage the YAML frontmatter required by SKILL.md files, ensuring the structured metadata is correctly formatted for Claude Code skill triggering and evaluation loops.

A Claude Code skill under-triggers or falsely triggers when its description lacks precision, which you can fix by running benchmark evaluations on example query sets and applying iterative description optimization.

The best way to iteratively refine a skill description is to run evaluation benchmarks across query variants, analyze performance patterns, and apply targeted description optimization to correct trigger failures.