skill-creator

To create Claude skills, you can draft SKILL.md frontmatter and body, run automated trigger-evaluation loops, collect performance metrics, and compare iterations to refine the skill. This scaffolding and testing process ensures safe, deployable automation workflows.

Automated skill evaluation works by running trigger-eval loops that benchmark performance across iterations. The system collects metrics, compares cycles, and surfaces actionable improvements for descriptions, scripts, references, and assets to optimize triggering and execution.

Skill creation requires the yaml dependency for frontmatter validation. The environment must also support CLI execution for scripts and component discovery to validate dependencies, check components, and ensure safe skill deployment.

Refining existing skills is fully supported. You can iterate on SKILL.md files, run new evaluation loops, collect updated performance metrics, and compare new iterations against previous cycles to improve triggering and execution.

Before deployment, skills undergo frontmatter validation, dependency checks, component discovery, and toxicity assessment. These automated checks ensure the skill is safe, structurally sound, and ready for production automation workflows.

Inconsistent performance metrics often stem from poorly drafted SKILL.md frontmatter or body content. Running automated trigger-eval loops and analyzing benchmarks helps surface actionable improvements to stabilize execution and triggering.