Eight people. One model.
One log everybody can read.
A shared model account turns a team into an anonymous crowd: nobody knows who asked for what, and whatever the model wrote lands in the repo unread. TagTeam gives the room a single append-only log — so every invocation carries a name, every change waits for a person, and the budget is spent on what the room pinned rather than on scrollback.
Nothing reaches the model by accident
Plain messages stay between you and cost nothing. Only a slash command spends budget, and the room sees a one-line notice rather than your tokens.
Nothing reaches main without a person
The model proposes; it never writes. Contract checks run on a scratch copy and the merge gate is enforced in the room, not in the button.
The ceiling is review, not GPU
The room measures how fast it generates against how fast it approves — and when generation runs away, it throttles itself.