A team runs in order: each agent builds on what came before. That is efficient, and it has an obvious weakness — if one agent does thin work, or is missing something, everything after it inherits the problem. Three settings on a team change what happens at those moments. They are independent; turn on whichever match how your team actually fails.
Let agents ask each other
Turn this on when your outputs contain confident numbers nobody can source. An agent that is missing a figure a teammate should have will otherwise assume one. With this on it asks by name, the teammate answers that one question, and the asking agent redoes its work with the answer. You see the question and the answer on the run.
Send weak work back for revision
A reviewer already critiques every multi-agent run. With this on, that critique is acted on: an agent judged materially deficient is sent back once, with the specific criticism, to redo its own work — before the run is called done. At most two agents per run, so nothing spirals.
Turn this on for work you send to someone else: a board paper, a client deliverable, anything where a weak section costs you more than the extra minutes.
Let the team bring in a specialist
You pick a roster before you know what the work will turn up. With this on, the coordinator may add one more agent at the end if the finished work revealed a gap the roster could not cover — a tax specialist because the diligence hit a cross-border structure, say. It only recruits on a concrete gap, never to polish what is already done.
What costs what
Each setting costs extra agent invocations, and only when it triggers: a question costs two (the answer, then the re-run), a revision costs one per agent, a recruited specialist costs one. A run with none of them triggering costs exactly what it does today.
Where to set them
All three are on the team builder when you create a team, and on the team's Edit page afterwards. They are per team, so a team with a deliberately fixed roster and a tight budget can stay exactly as it is.