skill-creator

To create and evaluate Claude skills, draft the skill, run automated evaluations to measure trigger accuracy and output quality, then iterate based on benchmark results until performance stabilizes. This workflow consolidates tooling, references, and assets into a portable skill package for deployment and reuse.

Automated skill benchmarking measures trigger accuracy, latency, and output quality during Claude skill development. You need it when iterating on skill drafts to ensure predictable performance in real prompts, governing an improvement loop that refines trigger descriptions and response quality.

To bundle scripts, references, and assets into a portable skill package, consolidate these components during the creation workflow. This packaging enables deployment and reuse of Claude skills across different environments, ensuring all necessary resources travel with the skill definition.

Yes, you can modify existing Claude skills using this workflow. It supports capturing intent to draft new skills or alter current ones, defining evaluation criteria, running automated benchmarks, and iterating on trigger descriptions to improve performance for real prompt scenarios.

Yes, pyyaml is required as a dependency to run the automated evaluation and benchmarking workflows. It supports parsing configuration files that define evaluation criteria, trigger accuracy tests, and skill packaging operations throughout the iterative refinement loop.

Inaccurate Claude skill triggers occur when trigger descriptions lack refinement. Use automated evaluation and benchmarking to measure trigger accuracy, then iterate on the trigger description optimization process until results stabilize and the skill activates predictably.