Agent skills have multiplied faster than any place to find them: a SKILL.md usually lives in a scattered gist or repo, discovered by word of mouth. The bet here is that skills deserve the same publish-version-discover loop npm gave code — a canonical slug, changelogs, and search that understands what a task needs rather than which keywords it spells.
What Sets It Apart
- Search runs on embeddings (OpenAI
text-embedding-3-smallover Convex vector search), so you find a skill by describing the job instead of guessing exact terms. - Canonical slugs support rename and merge while keeping old links and installs working through redirect aliases — maintainers can reorganize without breaking anyone's setup.
- Install pinning freezes a local copy so updates and force-reinstalls can't silently overwrite it, which matters when a skill is load-bearing in your workflow.
- One catalog spans both text skills and native code/bundle plugins, reachable from the same CLI (
clawhub package explore,clawhub install), so consumers don't juggle two ecosystems.
Who It's For
Great fit if you build or consume OpenClaw skills and want a versioned, discoverable home with moderation and star/comment curation. Look elsewhere if you're outside the OpenClaw ecosystem — the registry, CLI, and package format are OpenClaw-specific (code-plugin manifests must declare openclaw.compat.pluginApi), and the project is young, launched early 2026 on a Convex + GitHub-OAuth stack.
