About SkillRepo

Every programming ecosystem eventually builds the same infrastructure: a standard format, then a distribution layer — and, once teams depend on it, governance over what runs where. JavaScript got npm. Containers got Docker Hub. Python got PyPI. AI agent skills are following the same path. SkillRepo is building the distribution and governance layer for them.

The moment

In 2023, developers used AI chat for one-off tasks. Prompts were disposable. By 2024, teams were writing reusable skill files — structured instructions that made their agents genuinely useful. These lived in local directories, got shared in Slack, copy-pasted between repos. The best agent expertise was trapped on individual laptops.

The agentskills.io specification changed the format problem. A single markdown file with structured metadata gave skills a portable shape that runs in Claude Code, Cursor, Windsurf, Gemini CLI, Codex CLI, Cline, and VS Code + Copilot. Remote agents — and desktop apps like Claude Desktop (Cowork) and ChatGPT — can also connect via the MCP endpoint as a custom connector (connect an MCP client). But a standard without distribution is only half the story. Skills still had no central home — no way to publish, discover, or stay current.

That’s the gap SkillRepo fills.

Our approach

SkillRepo is the open distribution and governance layer for the agentskills.io specification. We don’t define the format — the open spec does. We make it usable at scale and accountable in production. Publish a skill once and every team that installs it stays current. Teams give each repo the skills it needs, pin the version it runs, and see any drift. Progressive disclosure means agents load only what they need: lightweight metadata at startup, full instructions on activation, supporting files on demand. One install command, zero maintenance.

Safety signals are surfaced on every skill. Every skill published to SkillRepo is automatically analyzed by AI for safety, quality, clarity, and completeness — graded A through F to help you make an informed call. Skills run inside agents with access to your filesystem, terminal, and code, so the more you know before you activate one, the better. Grades are signals, not a certification; always review a skill yourself before use.

What we believe

• Skills should be portable. Your instructions belong to you, not to a vendor. The agentskills.io spec is open, vendor-agnostic, and maintained by the community. No proprietary formats, no walled gardens.
• Standards matter. We didn’t invent a format — we built on one. The agentskills.io specification is the source of truth for how skills are structured. SkillRepo conforms to it completely.
• Teams need a shared source of truth. Private libraries, approved-version pins, and drift you can see let organizations govern their agent knowledge centrally — not scattered across repos and config files.
• Transparency over trust. Skills run inside AI agents with real access to your tools. Every skill in the catalog is automatically analyzed for safety signals, quality, and completeness. We don’t claim to catch every issue — we surface what we find so you can make informed decisions before activating anything.
• Community makes everyone better. The best skills shouldn’t live on one person’s laptop. Public skills let developers share what works and build on each other’s expertise.

Get in touch

We’d love to hear from you. Reach us at hello@skillrepo.dev.

Questions