Create, test, and optimize Claude skills with structured evaluations and benchmarks.
npx skills add https://github.com/yamcodes/arkenv --skill create-skill-yamcodes
Please help me install this Agent Skill. Skill: create-skill Source: https://github.com/yamcodes/arkenv/tree/main/skills/create-skill Command: npx skills add https://github.com/yamcodes/arkenv --skill create-skill-yamcodes
This Skill guides teams through creating new Claude skills, refining existing ones, and measuring skill performance with automated evaluations and benchmarks. It enables rapid iteration from concept to triggering-ready descriptions, reducing drift between intent and behavior.
Draft a new skill, define an initial eval plan, run iterations, and optimize the triggering description.
High-intent search queries and answers about installing and using this skill.
To create and optimize Claude skills end-to-end, you draft a skill, define an initial eval plan, run iterations, and refine the triggering description. This workflow reduces drift between intent and behavior until the skill performs reliably.
Skill benchmarking uses structured evaluations to compare trigger accuracy, performance, and efficiency across iterations. The eval-viewer coordinates these structured prompts to produce repeatable benchmarks for measuring skill development progress.
Yes, PyYAML is required as a dependency for this skill development workflow. The environment needs PyYAML installed to coordinate optional resources like scripts, references, and assets during creation and evaluation.
The best way to refine triggering descriptions is through automated evaluation and benchmarking. You iterate on the description, run structured evals to measure trigger accuracy, and optimize based on results to improve discoverability across domains.
Skill behavior drifts from its original intent when triggering descriptions are not optimized. Running structured evaluations and iterating on the description until benchmarks perform reliably reduces this drift between intent and behavior.
Yes, you can use this workflow to test existing skills. It guides teams through refining existing skills and measuring performance with automated evaluations and benchmarks, not just creating new ones from scratch.
openoctaGenerate OpenOcta Skills with SKILL.md and resource directories.
TheBushidoCollectiveExport TensorFlow models to SavedModel and convert to TFLite for deployment.
lucasfthDesign, validate, and iteratively improve Claude Skills with SKILL.md and eval loops.
renatoasseDraft SKILL.md files and run evals to iterate Opensquad skills.
