Skills Manager gives you a single, governed place to centralize agent skills, control who can use them, and serve them securely to Claude and any MCP client — all synced from Git.
Connect GitHub repositories of skills and the hub syncs them automatically. Git stays the source of truth — you simply curate which skills from each source are published. No copy-paste sprawl, no laptop-only workflows, no version drift.
Assign each skill org-wide, to a specific team, or to an individual user. Access is least-privilege by default, so people and agents only ever see what they have been granted. Change grants at any time from the web UI or via MCP tools.
Use the OAuth 2.1 custom connector in claude.ai and Claude Desktop, or mint personal access tokens for Claude Code and Codex. Crucially, you can scope a token or connection to specific skills — for example, give one automated agent only three skills.
Users create their own skills directly in the hub. Every change is automatically pushed to their own Git repository, so individual work is versioned and backed up — never trapped on a single machine.
Feedback and edit proposals route to admins for review. Approved changes are committed back to Git. You get a clear approval trail for every modification to a shared skill — governance, not chaos.
A complete web admin UI covers sources, access, teams, tokens and review. The same operations are exposed as 40 MCP tools, so you can add a source, grant a skill, or mint a token without leaving Claude.
Mark skills as favorites and they are preloaded into the agent's context. Frequently used workflows are immediately available instead of having to be discovered each session.
Supabase Auth, row-level org isolation, Vault-encrypted GitHub keys, and a cloud deployment on Supabase + Vercel. Every organization's data is isolated at the database row level.