Create and iteratively improve Claude-compatible skills with automated evaluations.
npx skills add https://github.com/vobase/vobase --skill skill-creator-vobase
Please help me install this Agent Skill. Skill: skill-creator Source: https://github.com/vobase/vobase/tree/main/.agents/skills/skill-creator Command: npx skills add https://github.com/vobase/vobase --skill skill-creator-vobase
Helps developers and product teams turn informal workflows or ideas into robust Claude-compatible skills by guiding skill drafting, test creation, automated evaluations, and iterative improvements so skills trigger reliably and produce measurable outcomes.
Tell the assistant what you want the skill to accomplish, ask it to draft SKILL.md and a small eval set, then run an evaluation loop to compare with a baseline and produce a review.
High-intent search queries and answers about installing and using this skill.
To create and iteratively improve AI skills, draft a SKILL.md with metadata and triggering guidance, generate test cases, run with-skill versus baseline evaluations, grade outputs, and aggregate benchmark statistics to measure improvements.
The best way to benchmark AI assistant skills is to run parallel with-skill and baseline comparisons using generated eval sets, grade the outputs automatically, aggregate the resulting metrics, and review the statistics in a human-friendly HTML viewer.
Description optimization for Claude-style agent skills works by optionally calling the anthropic Python SDK to propose description revisions, running paired evaluations, and iterating until trigger accuracy measurably improves.
Yes, you can use the anthropic Python SDK to automate skill authoring and testing by generating eval sets, calling the SDK for description improvement, grading outputs, and aggregating benchmark statistics for validated skills.
Yes, pyyaml is required to generate test cases and run automated evaluations, as it handles the YAML metadata parsing needed for SKILL.md files, test-case generation, and benchmark aggregation.
A Claude skill may not trigger reliably if its description lacks clear metadata, so you should run description optimization, compare with-skill and baseline outputs, and iterate revisions based on quantitative grading results.
vuanhtu1993Create, validate, and package Claude Skills into .skill files.
taisukeoeAutomate evaluation-driven development for agent skills across Claude-based pipelines.
insmtxAuthor, validate, and optimize Claude Code skills with Python and YAML.
ArogyaReddyAutomates LLM output evaluation with metrics, human feedback, and benchmarking.
