skill-creator

To create a Claude Code skill, you draft a SKILL.md file with YAML frontmatter defining the name and description, then run automated evaluations to test trigger accuracy across generated prompts.

Optimizing Claude skill descriptions involves iterating on the YAML frontmatter text and running benchmark-style comparison loops. This process measures invocation accuracy to ensure the skill triggers only at the appropriate times.

Trigger-focused evaluation in prompt engineering is the process of generating a test set of queries and running automated eval loops to measure when a skill is or isn't invoked. It validates that description metadata correctly targets user intent.

Yes, you need the pyyaml dependency installed to parse and encode the YAML frontmatter required for skill creation. This metadata defines the trigger conditions evaluated during the benchmarking process.

Yes, you can perform a description-optimization pass on an existing skill to improve its triggering behavior. This involves analyzing quantitative summaries from eval runs and refining the trigger conditions without a full rewrite.

A Claude Code skill triggers incorrectly when its YAML frontmatter description lacks accuracy or scope. Running trigger-focused evaluations and analyzing post-run patterns helps identify and fix these invocation mismatches.