skill-creator

To create Claude Code skills, draft a SKILL.md with front-matter metadata and body instructions, then run automated trigger evaluations and benchmarks to iteratively improve triggering accuracy across varied user prompts.

Evaluating prompt triggering accuracy involves running automated trigger evaluations, spawning baselines, and comparing configurations to measure performance, stability, and improvement over iterations using analytics dashboards.

Yes, the anthropic dependency is required for running Claude Code interactions and trigger evaluations, while yaml is needed to parse and generate the front-matted SKILL.md metadata structure during skill creation.

Structured interviews extract user needs, edge cases, and success criteria during skill creation, ensuring the resulting SKILL.md instructions and trigger evaluations align with actual user intents rather than assumptions.

Yes, you can benchmark Claude skill performance by producing analytics and dashboards that track triggering stability and improvement metrics across multiple evaluation iterations and configuration changes.

Claude skills fail to trigger correctly when the SKILL.md description lacks eval-driven iteration; running trigger evaluations against varied prompts and comparing baselines identifies description gaps to improve reliability.