skill-creator

To create AI skills systematically, you capture the initial intent, draft a SKILL.md file, generate test cases, and run parallel execution benchmarking to evaluate performance. This structured workflow ensures skills trigger correctly and perform reliably before production deployment.

Optimizing AI skill descriptions requires running an automated refinement loop that uses train/test splits to prevent overfitting. This process adjusts the description text iteratively to ensure the skill triggers reliably only for relevant user queries.

Assertion-based evaluation benchmarks AI skills by spawning parallel test runs and grading the generated outputs against predefined assertions. The system then aggregates these quantitative performance metrics to compare different skill versions.

Yes, the skill creation workflow supports cross-environment deployment for GLM Code, GLM.ai, and Cowork. It packages your finalized instructions and assets into a .skill file format for production-ready use across these platforms.

Yes, pyyaml is a required dependency for parsing and managing the YAML frontmatter used in SKILL.md files. You need it installed in your environment to properly author, evaluate, and package the skill components.

Blind comparison removes bias during skill iteration by presenting outputs from different versions side-by-side in an interactive review viewer. This ensures you select the best-performing iteration based on aggregated quantitative grading rather than subjective preference.