AIBestSkill

How AIBestSkill keeps the AI skills shortlist small.

See how AIBestSkill scans GitHub daily, keeps an approved shortlist, and weighs stars, growth, maintenance, SKILL.md proof, and source excerpts before install.

SKILL.md proofNew candidates need a real SKILL.md.
>500 starsNew candidates need more than 500 GitHub stars.
Adoption momentumAfter the gate, ranking looks at stars, growth, and maintenance.
Review gateDaily discovery finds candidates. Review and QA decide what becomes public.

Why not a big directory

AIBestSkill is a trusted shortlist for AI skills worth installing, not a grab bag of AI tools. The point is what we leave out: many repos get discovered, but only approved objects become public pages.

Why too many skills can make agents worse

Skills change how an agent reads instructions, picks tools, and spends context. Adding a weak skill can make the agent noisier. Adding ten weak skills can make it harder to know what the agent is following.

What gets listed

  • Public GitHub repos that behave like reusable agent skills for Claude Code, OpenClaw, Codex, or adjacent workflows.
  • New candidates must have an actual SKILL.md and more than 500 GitHub stars before they can enter the public approval path.
  • Approved pages need source excerpts, a visible rank signal, recent repo activity, and enough evidence for a real install review.

What we refuse to list

  • Generic AI tools with no reusable skill object.
  • Private, unverifiable, or marketing-only pages without public source proof.
  • Repos that only look hot because of a topic, README wording, or one install command but have no real SKILL.md.

How ranking works

The default rank signal uses GitHub stars, measured star growth, and recent maintenance. Stars give an adoption baseline. Growth shows current momentum. Maintenance keeps stale repos from floating up forever. It is not a safety score.

Why stars, updates, and growth matter

  • Stars are imperfect, but they show whether enough developers have noticed the repo to make it worth a closer look.
  • Recent updates matter because skills sit inside agent workflows. A repo that has not moved in a long time may still work, but it should not rank like an actively maintained one.
  • Growth trend catches the difference between old popularity and current demand. A smaller repo gaining stars quickly can deserve review before a larger repo that has gone quiet.

How channel tags work

Claude Code, Codex, OpenClaw, and AI Agent tags are search and workflow channels. They help people find nearby skills, but they are not compatibility certification. Check the upstream repo before installing.

How evidence works

Detail pages show SKILL.md source proof, selected source excerpts, optional install notes, and related alternatives. Install notes are review aids. They are not a promise that every skill is one-click install-ready.

Daily discovery vs approved public shortlist

The scanner can find candidates every day. Public listing is slower on purpose. Review, the approved manifest, sitemap checks, and QA decide what can be indexed.