One skill library · every agent

Your team's know-how, in every AI agent

SkillRelay turns how your company works into living skills for Claude, ChatGPT, Cursor — anything your team uses. Anyone can fix a skill by saying what's wrong, in plain English, in a normal chat. The whole company gets it in minutes.

No password, no credit card — we email you a sign-in link.

claude.ai — Casey (Customer Success)
The file-a-bug skill points at the wrong Linear team — it should be Platform, not Core.
▸ read_skill(file-a-bug) · propose_change(…)
Updated file-a-bug to v4 — routing bugs to Platform. This reaches all 23 people at Acme within about 30 minutes.
Claudev4auto-synced
ChatGPTv4connector
Cursorv4AGENTS.md
Claude Codev4plugin
Works where your team already works
Claude · ChatGPT · Claude Code · Cursor · GitHub Copilot · Windsurf · Gemini CLI · Codex

Built for the people who actually know the work

The person who knows the bug-filing process is rarely the person who can edit a Git repo. SkillRelay removes everything between knowing and shipping.

Edited by talking

“The bug-filing skill points at the wrong team.” That sentence, said to any connected chat, is the entire editing workflow. No files, no uploads, no tickets.

Versioned forever

Every change is a new immutable version with a named author. Anything reverts with one sentence — safety through undo, with optional approvals when you want a gate.

Zero admin treadmill

One five-minute setup, once. Every future skill and edit reaches the whole org automatically — no admin ever touches it again.

How it works

1
An admin connects the workspace
Guided checklist, about five minutes, once. We host the plumbing — your IT team never gets a ticket.
2
Everyone signs in once
One click connects each teammate's Claude, ChatGPT, or editor — so every edit carries its author's name.
3
The library keeps itself current
Notice something stale? Say so. Skills version, publish, and reach every agent in the company within about 30 minutes.

Boring on purpose, where it counts

Least privilege

Our GitHub app can touch exactly one repo — contents and pull requests, nothing more. No admin scopes, ever.

Attributed & reversible

Per-person OAuth, tokens hashed at rest, every version immutable, one-click revert and offboarding.

Roles & approvals

Admins, Editors, Readers — plus optional review gates on the skills that matter. Trust dials, not bottlenecks.

The full security & privacy story →

Questions teams ask

Do my teammates need GitHub?

No. One admin links a GitHub username during setup — everyone else just talks to their AI tools. The repo is plumbing we host; nobody ever opens it.

We use ChatGPT and Cursor too. Does this help us?

Yes — that's the point. The same library serves Claude (auto-synced), ChatGPT (connector, deployable org-wide on Enterprise), and every MCP-capable editor. One fix updates them all.

What if someone breaks a skill?

Nothing is ever overwritten. Every change is a new version with a named author, and anyone can say “put it back how it was.” Optional approvals add a review gate for the skills everyone gets.

Where does our content live?

In an isolated, private store per workspace — encrypted volumes, hashed credentials, no analytics, never used for anything but serving your team. Details here.

What does it cost?

Free while in early access. We'd rather have your feedback than your credit card.

Stop re-explaining how your company works

Set it up in five minutes. The first time a teammate fixes a skill from their phone, you'll get it.