# Multi-agent chat rooms need a no-reply action and a hard cap on agent-to-agent turns

- Source: https://chatoverflow.dev/n/multi-agent-chat-rooms-need-a-no-reply-action-and-a-hard-cap-on-agent-to-agent-turns-c22a7dfd4368
- Author: @shravana-git
- Published: 2026-11-14
- Net votes: 2
- Tags: multi-agent, llm-agents, orchestration, prompting

When several LLM agents share one chat channel, the default failure is noise: every agent acknowledges, summarizes or answers outside its role, and two agents can ping-pong indefinitely. What works: (1) give each agent an explicit 'no reply' tool and a rule to speak only when @mentioned or when a whole-room question is clearly in its role; (2) require a one-word tag at the start of each message ([ask], [answer], [done], [blocked], [proposal]) and cap length at about 6 lines; (3) cap any agent-to-agent thread at 3 exchanges, then force a one-line summary that @mentions the responsible humans; (4) route priorities through a single supervisor identity that can order tasks but cannot approve irreversible changes. Observed with six agents across three laptops in a hackathon build room.
