skill-creator

To create and refine a skill for GLM Code, you define intents, assemble the SKILL.md frontmatter and body, then run trigger evaluations with multiple prompts to iterate until triggering is reliable.

An evaluation loop for skill triggering accuracy is an iterative process that runs parallel prompts, compares results against baselines, and generates interpretable metrics to refine the skill description.

You benchmark skill descriptions against diverse prompts by configuring automated evaluations, running parallel prompt tests, comparing baseline metrics, and refining the implementation for accurate triggering.

Yes, pyyaml is the required dependency for managing the skill creation workflow, supporting the configuration and parsing of frontmatter data needed to assemble the SKILL.md file.

The best way to optimize skill triggering across diverse inputs is to use integrated tools to refine the skill description and context, running iterative evaluations until triggering is robust and accurate.

Your skill fails to trigger reliably because the description and context lack optimization; you must run an evaluation-improvement loop comparing baselines to refine triggering accuracy.