Run one workspace per client or matter

If your work is organized around clients, matters, or projects — the way law firms, consultancies, and agencies work — the cleanest way to keep them apart is to give each one its own workspace.

What a workspace isolates

A workspace is the hard boundary in Agentman. Each one has its own:

• Skills library — My, Team, and Shared-with-me shelves that exist only there.
• Members — who's in Client A's workspace has no bearing on Client B's.
• Permissions — every grant, including use-only, stops at the workspace wall.
• Activity logs — usage history is recorded and viewable per workspace.

Nothing crosses between workspaces unless someone deliberately exports a skill or clones it across. There is no accidental path from one client's library to another's.

Two grains of control

• Inside a workspace, per-skill access levels (use / read / edit / admin) are the fine grain — see sharing a skill.
• Between clients, the workspace is the wall. You don't manage cross-client confidentiality with per-skill grants; you manage it by never putting two clients in one workspace.

Connect them all to Claude

Claude allows one connector per URL, so we provide three identical endpoints — connect each workspace through its own:

https://skills.agentman.ai/mcp
https://skills2.agentman.ai/mcp
https://skills3.agentman.ai/mcp

Add each as its own custom connector, and during that connector's OAuth consent, sign in to the workspace it should serve. Claude then holds Client A's and Client B's libraries side-by-side — separate permissions, separate logs, no cross-contamination. Full steps in Connect your skills library to Claude.

Moving work between workspaces

When something should cross the wall — a generic playbook you want to reuse for a new client — that's a deliberate act: export the skill as a package and import it, or clone it across. Both leave the original untouched.

Availability

All plans. Each workspace has its own plan and billing.