peer-agentslisted

# Peer Agents Bring another AI into the current work as a visible, independent colleague — not an echo of the current session. The coordinator opens fresh AI chats in cmux, gives each a useful role, reads their answers from the terminal, follows up when needed, and returns its own judgment in the main chat. ## Why bring in peers The useful question is not *which model is best* — it is *which role is missing*. A second AI is worth the cost when its **separate line of reasoning** adds something the current session cannot produce on its own: - **Combine compute, don't pick one.** Two independent thought chains, merged with attribution, beat a single chain — one model's planning depth complements another's execution or breadth. - **Break the echo chamber.** A fresh session with no exposure to the coordinator's reasoning won't rationalize the coordinator's mistakes. - **Different model families fail differently.** A different family catches bugs, blind spots, and bad assumptions that the same family would repeat. - **No self-grading.** Independent review and verification are only meaningful when the reviewer is not the author. - **Cheap parallel exploration.** Several peers can chase alternative approaches at once, so you compare real options instead of imagining them. This skill gives you the *ability* and the *reasons*. You write the actual prompts and decide the use case. Read [references/tools.md](references/tools.md) before launching peers. It is the only op