skill-builder

Progressive disclosure in skill-building involves writing a lean SKILL.md supported by rich references and bundled resources like scripts. This approach manages complexity by revealing detailed information only when needed during the workflow.

You evaluate Claude Code skills by running trigger evaluations and comparing outputs using the provided tooling. This guided benchmarking process tracks performance and helps iterate toward better triggering and overall skill quality.

The best way to structure a SKILL.md file is using progressive disclosure with lean frontmatter and optional bundled resources. This keeps the core file lightweight while leveraging references and assets to handle workflow complexity.

Yes, you need Python dependencies including anthropic and pyyaml to run skill evaluations and benchmarks. These libraries enable the scripts to interface with Claude and parse configuration files during the testing loop.

Yes, you can iteratively improve skill descriptions to optimize triggering by applying an operator-style workflow. This integrates Push description optimization and analysis references to refine how skills activate over time.