teams-and-workgroups

Teams and rooms

For developers ready to compose multiple agents around a shared goal. Teams define who works together; rooms are where they do the work.

This concept used to be called a Workgroup. Activating a team now creates room-<N>-<team>. Any wg-* directory you already have keeps its name and keeps working exactly as before, and the CLI still accepts workgroup, purge-wg, --wg and --workgroup as deprecated aliases of room, purge-room and --room. A later release removes the aliases; nothing on disk is ever renamed.

Team

A team is one orchestrator plus one or more worker agents. The orchestrator and every member must already exist as agent matrices before you create the team. The team's config lives at .ac/_team_<name>/config.json and lists members by their canonical names.

my-project/
└── .ac/
    ├── _agent_tech-lead/
    ├── _agent_dev-rust/
    ├── _agent_dev-ts/
    └── _team_feature-x/
        └── config.json

config.json (simplified):

{
  "coordinator": "_agent_tech-lead",
  "agents": [
    "_agent_tech-lead",
    "_agent_dev-rust",
    "_agent_dev-ts"
  ],
  "repos": []
}

You create teams from the Teams UI in the sidebar, or from the CLI with team create. Pick an existing orchestrator agent, pick one or more existing member agents, optionally define repo access, then save. Rooms are created later when you activate the team for a task.

Orchestrator authority

The orchestrator is the only team member that can:

• Send messages to any other team member (members can only message peers in the same team plus their orchestrator).
• Edit the room TASK.md through the CLI (task-set-title, task-append-body).
• Close other members' sessions (close-session).
• See the synthetic agentscommander://root-agent peer when verified.

This is enforced at the daemon mailbox boundary. Non-orchestrator attempts return an authorization error.

One agent, many teams

The same agent matrix can belong to multiple teams. Each team that includes the agent gets its own replica when a room activates — replicas are independent working copies, so the agent runs separately in each team's room.