Same brief
Each proof starts with one real request, written once and never changed between runs.
An independent test for anyone’s skill: the same brief run three times without it and three times with it. Every result is a live page you can click through, free and in public.
Hover any proof to wipe across to the version Claude built without the skill. Every proof is open, three runs each way.
Star counts tell you a skill is popular. A proof shows what it actually changes, on a task like yours.
Each proof starts with one real request, written once and never changed between runs.
Claude builds it on its own and with the skill loaded, three independent runs per side on the same model, so you see the spread and not one lucky draw.
Both results are live pages. Drag the divider, click around, and judge the difference yourself.
Free to search and free to install. Every listing links straight to its source. New to skills? Read how to use one.
Search by what you're building, or browse by category.
⌕ Try “landing page”, “postgres” or “video”From 48 public repos, refreshed weekly
Each skill comes with a command that copies it into your project.
New skills are picked up and dead links drop out every week. Last refresh November 9.
Most directories rank skills by stars or installs. Those numbers say who found a skill, not what it does. Skill makers test their own skills; we test anyone’s, the same way every time, and publish every run.
What most lists rank by
Better, still indirect
What Use This Skill adds
Every listing is scanned for the nine patterns a skill should never need. Check any skill on GitHub before you install it, or run the same scan on your own repositories in one command or on every pull request.
CLI and GitHub Action, free forever.
Private repositories, one key for the organization.
A skill is a folder with a SKILL.md file that teaches a coding agent how to do one kind of job well, such as designing a page or writing tests. Claude Code, Codex, Cursor, OpenCode and other agents load them when the task calls for it.
We give Claude one fixed brief three times without the skill and three times with it, on the same model. Every result is saved as a live page, with the model and the skill's exact version recorded next to them.
Yes. The brief, model and skill version are all recorded, so anyone can run the same comparison again. We re-run proofs when a skill changes.
Everything on the site: searching and installing all skills, every proof with all three runs each way, the safety scan on every listing, the connector, checking any skill on GitHub before you install it, and scanning your own public repositories. Teams pay $29 a month to scan private repositories on every pull request.
Their authors do. We list each skill with a link to its source repository and only show a skill's full text when its license allows it.
Every listing has a one-line install command that copies the skill folder into your project's .claude/skills directory.
Every proof is open. Scan your own repository in one command.