Create, edit, and optimize Claude skills with evaluation workflows.
npx skills add https://github.com/Hashzin-0/Curion --skill skill-creator-hashzin-0
Please help me install this Agent Skill. Skill: skill-creator Source: https://github.com/Hashzin-0/Curion/tree/main/.opencode/skills/skill-creator Command: npx skills add https://github.com/Hashzin-0/Curion --skill skill-creator-hashzin-0
Reduces the friction of designing, testing, and iterating Claude skills by providing a structured, repeatable workflow for drafting SKILL.md, generating test prompts, executing paired with-skill and baseline runs, grading outputs, and aggregating benchmark results. It helps teams move from an informal prompt to a packaged, evaluated skill with clear quantitative and qualitative feedback.
Ask for help defining intent, draft SKILL.md, and produce 2-3 realistic test prompts plus basic eval assertions so the skill-creator scripts can run an initial iteration.
High-intent search queries and answers about installing and using this skill.
To create and test Claude skills, draft a SKILL.md file, generate realistic test prompts, and run paired with-skill and baseline evaluations using scripts like run_eval.py. The workflow captures timing, tokens, and grades outputs against assertions to produce quantitative benchmarking results.
Optimizing a SKILL.md description involves running an automated description-optimization loop that iterates on the skill's frontmatter and content. This improves discoverability and invocation rates by evaluating baseline and with-skill outputs, then refining the description for better triggering accuracy.
Yes, PyYAML is a required dependency for running skill evaluation and benchmarking scripts. It is needed to parse the YAML frontmatter in SKILL.md files and manage the eval_metadata.json configuration used by the automated testing workflow.
Yes, you can compare skill versions by executing paired with-skill and baseline runs. The aggregation scripts compute pass rates, time, and token deltas, allowing you to benchmark two versions and generate analyst observations and comparator analyses.
The best way to package skills for reproducible benchmarking is to bundle the SKILL.md, reusable scripts like run_loop.py and aggregate_benchmark.py, and skill assets into a distributable .skill file. This ensures all evaluation dependencies and test cases are contained.
s977043Design a Claude Code skill from a use case and produce a repo-ready package.
AndurilCodeEvaluate context engineering harnesses against no-harness baselines with graded assertions.
qiushibangCreate, modify, and improve AI agent skills through eval-driven workflows.
gesmith0606Create, evaluate, and iteratively improve SKILL.md descriptions with Python scripts.
