| name | team |
| description | Run a task as a multi-agent team — plan, have an independent model audit the plan, delegate execution to the right models, independently verify against the plan, integrate. Use when the user types /team, asks to "run this as a team", "use the team", plan-execute-verify, or wants work split across multiple models. |
Multi-agent team protocol
You (the current session) are the orchestrator. You run a
plan → audit → execute → verify → integrate loop using the team roster.
Step 0 — Read the roster (always, first)
cat ~/agent-team/team.json
The roster is the single source of truth for who is on the team and who does
what. Never hardcode member names or models — the user edits this file as
models are added and removed. Honor enabled flags, the assignment map,
and policies (fallback rules live there). If the roster is missing or
unparseable, say so and fall back to working solo.
Step 1 — PLAN (orchestrator)
- Decompose the task into self-contained subtasks; classify each
hard / default / trivial (this picks the executor tier) and note the
proposed executor per subtask — the auditor reviews this tiering.
- For worker subtasks, prepare a scratch dir or git worktree and write a
SPEC.md into it: goal, constraints, acceptance criteria, how to test.
- Write the master plan to a
PLAN.md beside the specs (subtasks, proposed
tiers, dependencies between subtasks) — the plan auditor and the verifier
both need it verbatim.
Step 2 — AUDIT THE PLAN (per roster assignment.plan_auditor, gate before execution)
Before any executor starts, send the plan to the plan auditor — consult
mode, read-only, one channel per task. Invoke it with the auditor's
invoke.consult command from the roster, attaching :
