Governed control plane for AI skills
The governed control plane for your AI agents’ skills
Centralize your organization’s agent skills, control who can use which one — org, team, or person — and serve them securely to Claude and any MCP client, synced straight from Git. Stop letting workflows scatter across local laptops; make them consistent, access-controlled, and connected.
Works with Claude, Claude Code, Codex & any MCP client · Free to test for 30 days
The problem: skills are scattered and ungoverned
Teams accumulate reusable agent skills — workflows, SOPs, scripts — but they live on individual laptops and in private repos. Nobody knows what exists, versions drift, and there is no way to control which agent or person can run what. As AI agents spread across an org, that sprawl becomes a security and consistency problem.
The solution: one governed hub
Skills Manager makes Git the source of truth, publishes a curated library, and serves it over MCP with granular access control. Grant skills org-wide or to a single person, scope tokens to specific skills, and connect any AI client securely — managed from a web UI or 40 MCP tools.
Everything you need to govern agent skills
Centralized skill library
Sync your organization's skills from GitHub repos — Git is the source of truth. Curate exactly which skills from each source are published to the hub.
Granular access control
Grant each skill org-wide, to a specific team, or to an individual user. Least-privilege by default — people see only what they should.
Secure connections
OAuth 2.1 custom connector for claude.ai / Claude Desktop, plus personal access tokens for Claude Code and Codex. Scope a token or connection to specific skills.
Personal skills + Git backup
Users create their own skills in the hub; every change is auto-pushed to their own repo, so nothing lives only on a laptop.
Governance workflow
Feedback and edit proposals flow to admins for review, then get committed back to Git — controlled, auditable change management for skills.
Manage anywhere
A full web admin UI and 40 MCP tools. Do everything — add sources, grant access, mint tokens — from your AI client or the browser.
Favorites in context
Mark skills as favorites and they are preloaded into the agent's context, so the right playbooks are always one prompt away.
Security & multi-tenancy
Supabase Auth, row-level org isolation, Vault-encrypted GitHub keys, cloud-hosted on Supabase + Vercel. Built for teams that need isolation.
How it works
- 1
Connect a Git source
Point the hub at a GitHub repo of skills. The hub syncs them and you choose which ones to publish.
- 2
Control access
Grant each skill org-wide, per team, or per person — least privilege, no scattered local copies.
- 3
Connect your AI client
Add the OAuth 2.1 connector in claude.ai / Claude Desktop, or use a scoped personal access token in Claude Code / Codex.
- 4
Your agents use the skills
Agents discover and run exactly the skills you granted — favorites preloaded into context.
Who it’s for
Agencies & consultancies
Standardize playbooks and SOPs across every client engagement — one curated, access-controlled library instead of scattered local setups.
Ops, marketing & sales teams
Share curated AI workflows so everyone runs the same approved process — consistently and on-brand.
Enterprises
Get isolation, approval workflows, and control over exactly what your agents can do — with row-level org isolation and scoped access.
Secure and multi-tenant by design
- Row-level org isolation (Supabase RLS)
- Vault-encrypted GitHub keys
- OAuth 2.1 custom connector
- Scoped personal access tokens
- Git as the source of truth
- Least-privilege access
Frequently asked questions
What is Skills Manager?
It is a cloud control plane that centralizes your organization's AI agent skills, controls who can use which skill (org, team, or person), and serves them securely to Claude and any MCP client — all synced from Git.
What are agent skills?
Agent skills are reusable units of agent know-how — workflows, SOPs, scripts and instructions (an open standard at agentskills.io) that an AI agent can load and execute. The hub stores, governs and serves them.
What is MCP and why does it matter?
MCP (Model Context Protocol) is the de-facto open standard for connecting AI clients to tools and context. It is native in Anthropic, OpenAI, Google and Microsoft products and was donated to the Linux Foundation's Agentic AI Foundation in December 2025. The hub speaks MCP, so it works with any MCP client.
Which AI clients work with it?
Claude (claude.ai), Claude Desktop, Claude Code, Codex, and any other MCP-compatible client. Use the OAuth 2.1 connector for desktop apps or a personal access token for CLI tools.
How does access control work?
Every skill can be granted org-wide, to specific teams, or to individual users. You can also scope a connection or token to a subset of skills — for example, give one agent only three skills.
Is my code and data secure?
Yes. The hub uses Supabase Auth with row-level org isolation, Vault-encrypted GitHub keys, OAuth 2.1, and least-privilege access. Your skills stay in your own Git repos as the source of truth.
Do I have to migrate my skills?
No. Git stays the source of truth. You connect existing GitHub repos and choose which skills to publish — no lock-in, no copy-paste sprawl.
How much does it cost?
There is a Free tier for individuals and small orgs, a Team plan for shared, access-controlled libraries, and an Enterprise plan with SSO and isolation. See the pricing page for current, indicative tiers.
Bring your agent skills under control
Centralize, govern, and securely serve your skills to every AI client in minutes.