| name | team |
| description | Use when one execution lane is no longer enough and work should escalate to the durable `agmo team ...` runtime. |
| argument-hint | [task, worker split, or team-runtime request] |
Agmo Team
Use agmo team ... when lightweight in-session help is not enough and you need a durable multi-worker runtime.
Core posture
- the leader stays in the current session as orchestrator
- durable state under
.agmo/state/team/<team>/... is the control plane
- tmux panes are worker transport and UI, not the source of truth
- native subagents are still better for small bounded sidecars
When to use agmo team
Escalate to team mode when you need one or more of:
- multiple independent implementation / verification / planning / knowledge lanes
- durable monitoring beyond one reasoning burst
- shared task claiming, worker heartbeat, or worktree-backed execution
- separate implementation and verification lanes for completion-gated execution (
ralph / strict execute)
For small same-session sidecars, prefer native subagents instead.
Leader responsibilities
The leader should:
- keep the task brief current and scoped
- start the team with
agmo team start ...
- verify startup with
agmo team status <team>
- monitor with
agmo team status, agmo team monitor, and agmo team hud
- prefer reclaim/rebalance/cleanup over ad hoc worker intervention when lanes stall
- own final integration, verification, and completion reporting
HUD and layout operations
Default to durable state before pane UI: agmo team status <team> and .agmo/state/team/<team>/... are the control plane.
Use agmo team hud <team> --preset sidecar when a live tmux sidecar is useful. The HUD also supports:
--preset focused for operator triage with worker diagnostics and read-only/dry-run-safe command hints
--preset full for deeper worker/task diagnostics and manual action review
--watch for a refreshing terminal view; use --refresh-ms and --iterations for bounded watch loops
If the sidecar prints inspect=..., treat it as a read-only hint first. Check worker/task state or run
agmo team layout status <team> before attempting layout changes.
For layout work, inspect before mutating:
agmo team layout status <team>
agmo team layout repair <team> --dry-run
agmo team layout rebalance <team> --dry-run
Only run repair or rebalance without --dry-run after the status/dry-run output justifies it. Non-tmux teams may
return skipped/not-configured layout results; do not infer that a live tmux session is required for durable team state.
Runtime artifacts to trust
Useful runtime artifacts live under .agmo/state/team/<team>/, including config, phase, tasks, worker status, heartbeat, inbox, mailbox, and dispatch state.
If tmux UI and durable state disagree, trust the durable state first.
Lifecycle
- start the team with a scoped task
- monitor progress from runtime state, not guesswork alone
- integrate and verify once worker lanes report completion
- shut down explicitly with
agmo team shutdown <team> when the work is done or aborted
Native subagent lifecycle
For any small native-subagent sidecars used before or around team escalation, keep each agent id until its result is integrated, then call close_agent for completed, failed, superseded, or no-longer-needed lanes so thread slots are released before the next delegation.
Artifact save body
Before shutting down or reporting completion for a team workflow, make the leader's final report save-ready:
- team task and worker allocation
- worker outputs integrated
- changed files or artifacts produced
- verification evidence
- remaining risks and follow-up ownership