Docs · Bundles
Bundles
A bundle is a named, versioned set of skills — like a playlist, but for your agent. Collect the skills your team uses every day, share them across machines, and hand a scoped key to a client so their agent can pull the same set.
What a bundle is
A bundle is a manifest — a list of skill slugs at pinned versions — plus optional metadata (name, description, visibility). When an agent syncs against a bundle, it installs exactly the skills listed, at exactly the versions pinned, and nothing else.
You create and manage bundles from the /library dashboard. Free accounts get 2 private bundles, Pro gets 50, and public bundles are unlimited on every tier including Free. See pricing.
Public vs private bundles
Public
Anyone with the bundle URL can install it. Useful for sharing a curated skill set with the community or documenting your team's standard toolkit publicly.
Private
Accessible only via an API key you explicitly share. Clients or teammates install with your key — the bundle content is never publicly discoverable.
Personal vs org-scoped (Pro)
Personal bundles belong to your account and sync to the machines where you run agents with your API key.
Org-scoped bundles (Pro) belong to your Pro account. Each client gets its own bundle-scoped API key, so the agent holding it can only reach the skills you chose — nothing else in your private catalog.
Pre-built catalogs for clients (Pro)
The canonical use case for Pro bundles: you operate AI agents for clients. Instead of manually installing skills on each client's machine, you:
- 1 Build a bundle in your library — pick the skills, pin the versions.
- 2 Assign a dedicated API key scoped to that bundle.
- 3 Give your client the one-line meta-skill install command that includes that key.
- 4 The client's agent installs the meta-skill once, then reconciles against the bundle each time its cron fires.
Edit the bundle — add a skill, bump a version — and every agent holding that key picks the change up on its next fire. You do that once, for all of them.
The direction matters: the agent asks, LoopSkill answers. Nothing here reaches a client's machine, so there is no per-agent "applied" state to watch and no failure for you to retry. The delivery walkthrough spells out what that means in practice.
White-label guidance
The client-side experience never mentions LoopSkill branding unless you choose to include it. The install command, the skills, the skill output — all run locally on the client's machine. No LoopSkill UI, no LoopSkill domain in their agent's context. You bill them under your own brand at your own rate.
Ready to hand one to a client?
The delivery walkthrough covers the key, the install command, and what you can and cannot see afterwards.