Write how you work once. Every agent follows.
Skills are your approved ways of doing work, written down for AI: how you answer a refund request, close the month, write in your brand's voice. connect0 keeps one tested, reviewed copy of each, shares it team by team, and serves it to connect0's agents, Claude Code, Cursor, Codex, GitHub Copilot and any MCP client.
In beta and free for now: versions, review, access, distribution and revoke carry no charge. Agents running a skill, and tests judged by a model, use credits like any other model call.
Checked before it goes live.
- 01
Written in the standard format
An Agent Skills folder, the format Claude, Codex, Copilot and Cursor already read. Write one in connect0, upload a folder or import from GitHub; it exports back unchanged. - 02
Tested on real examples
Test cases with an input and the check its answer must pass, exact or judged by a model. The reviewer sees the pass rate before approving. A skill that can change things runs its tests only when a person confirms. - 03
Reviewed, then published
With review on, a change stays a draft until a reviewer approves it. Every version is kept: roll production back with a reason, or pin an agent to a version. - 04
Shared team by team
Decide per team, person or agent who may use, change or review each skill. Every change, approval and use is in the account's audit log.
In the tools your people already use.
connect0 agents and Zero
They use the production version the moment it changes, checked against the person who started the run.
What connect0 seesEvery use, by version.
Any MCP client signed in to connect0
Listed live over the MCP Skills extension, and each skill as a prompt, filtered to what that person may use.
What connect0 seesEvery prompt use.
Claude Code, Cursor, Codex and GitHub Copilot
Installed from a private GitHub repository that connect0 writes and keeps current: one plugin per team, with the admin settings to paste into each tool.
What connect0 seesNothing: these tools do not report use back.
AWS Agent Registry, Google Agent Registry (preview) and Claude API skills
Signed copies pushed into your company's own catalogues with your credentials, and removed when you revoke a version.
What connect0 seesNothing: registries do not report use back.
connect0 counts what runs through connect0. A copy installed in another tool runs there, and that tool does not report back; the skill’s usage panel says so too.
Take it back everywhere.
Revoke a version with a reason. connect0’s agents, Zero and the MCP server stop serving it at once. The marketplaces and registries drop it on the next sync, minutes later, and connect0 lists any step left to do by hand in another tool’s admin settings. Put the previous version back in production in the same action.
Every production version is signed with your account’s key, and only signed versions leave connect0. Most other tools do not check skill signatures yet, and connect0 does not pretend they do.
Refund requests
Version 7 · in production for Support
- Tests
- 12 of 12 passed before approval
- Reviewed by
- Head of Support
- Delivered to
- connect0 agents and Zero · MCP server · acme/ai-skills (Claude Code, Cursor) · AWS Agent Registry
- Uses this week
- 312 through connect0 · other tools do not report
Revoke version 7
Copies drift. One source does not.
| Feature | Copied into each tool | connect0 company skills |
|---|---|---|
| Control | ||
| One approved version of each skill | No | |
| Tested and reviewed before a change goes live | No | |
| Who may use, change or review it, per team | per tool | |
| Reach | ||
| In the AI tools your people use, without copying | No | |
| A bad version withdrawn everywhere at once | No | |
| Which version ran, and how often | No | through connect0 |
connect0’s MCP server serves skills over the MCP Skills extension. Its public server card says so, with the skill tools beside it.
Run this in a terminal.
curl -s https://mcp.connect0.ai/.well-known/mcp/server-card.json | grep -oE '"(io\.modelcontextprotocol/skills|skills\.[a-z_]+)"'The extension id, io.modelcontextprotocol/skills, and the tools an agent uses to export, upload, test and track skills.