skill-creator

To create and evaluate GLM Code skills, this skill orchestrates an end-to-end workflow that drafts initial concepts, runs iterative evaluation loops with train/test splits, and generates structured outputs like SKILL.md updates and benchmark reports to refine triggering accuracy.

Benchmarking GLM Code skills involves running automated evaluation loops that measure performance, generate benchmark reports, and apply automatic description optimization to improve triggering accuracy based on user feedback and results.

You can set up train/test splits during the iterative evaluation loops to validate skill performance. This process structures evaluation prompts and provides result-driven updates to ensure reliable triggering before deployment.

Yes, PyYAML is used to manage structured outputs within the skill creation workflow. It facilitates parsing and generating SKILL.md updates, evaluation prompts, and benchmark reports for iterative refinement.

Yes, the skill creation workflow supports automatic description optimization. It orchestrates iterative evaluation loops that analyze results and automatically refine skill descriptions to improve triggering accuracy.

Iterating on skill creation using an end-to-end workflow solves the problem of unreliable triggering. It provides a repeatable process to draft, benchmark, and refine skills until evaluation results demonstrate consistent performance.