Create, test, and optimize AI agent skills with structured evaluation workflows.
npx skills add https://github.com/UniFormTS/UniForm --skill skill-creator-uniformts
Please help me install this Agent Skill. Skill: skill-creator Source: https://github.com/UniFormTS/UniForm/tree/main/.agents/skills/skill-creator Command: npx skills add https://github.com/UniFormTS/UniForm --skill skill-creator-uniformts
Building effective AI agent skills requires more than just writing instructions—it demands systematic testing, measurement, and iterative refinement. Without a standardized workflow, developers struggle to know whether their skills actually trigger correctly, perform better than baseline, or generalize across diverse user queries. This Skill eliminates that guesswork by providing a complete toolkit for creating, evaluating, benchmarking, and optimizing skills with quantitative rigor.
Use the skill-creator skill to build a new skill by describing what you want it to do, then run the evaluation loop to test and refine it until it passes all benchmarks.
High-intent search queries and answers about installing and using this skill.
To test and benchmark AI agent skills systematically, you can automate structured evaluation workflows that run parallel with/without-skill comparisons, aggregate pass rates, and grade quantitative assertions to measure performance against baselines.
To optimize skill descriptions for reliable triggering, you test and refine them against realistic query sets to maximize triggering accuracy while avoiding overfitting to specific phrasing.
Yes, you need Python to run Claude Code skill evaluations because it is required for benchmark aggregation, quantitative assertion grading, and generating the evaluation viewer interface.
The best way to build custom Claude Code skills is an iterative loop: capture initial intent, author the SKILL.md file, design test cases, run automated evaluations, and refine the implementation until passing all benchmarks.
Yes, you can run parallel subagent executions during AI agent skill evaluations to measure performance against baselines, visualize results in a review interface, and identify flaky assertions in benchmark data.
Custom skills fail to trigger correctly across diverse user requests when descriptions are overfitted, which you can resolve by applying automated description optimization against realistic query sets to maximize triggering accuracy.
coreyjaDraft SKILL.md frontmatter and structure guidance for Claude Code skills.
akbarfarooq2006Automates creation, testing, and refinement of Claude Code skills with built-in evals and benchmarking.
dparedesiCreate, evaluate, and improve Claude Skills using templates and scoring rubrics.
dnkrowValidate Claude Code skill directories for structural compliance and behavioral correctness.
