peer-agentslisted

# Peer Agents Bring another AI into the work as an independent peer, not as an echo of the current session. The value comes from assigning the right role, preserving a separate line of reasoning, and returning useful disagreement with clear provenance. The current session is the coordinator. It decides what perspective or contribution is missing, gives each peer only the context needed for that role, supervises the run, and makes the final judgment. A peer never becomes the uncredited source of the coordinator's answer. Read [references/tools.md](references/tools.md) before setup or execution. It is the sole operational reference for choosing models and modes, operating cmux, Claude Code, Codex, and Cursor, managing worktrees and sessions, and installing workflow rules. ## Operating principles 1. **Role is not model.** Define the contribution first—critic, alternative planner, reviewer, implementer, verifier, researcher, or another bounded role—then choose a suitable model and tool. 2. **Independence is deliberate.** Fresh peers do not receive other peers' answers. Do not preload the coordinator's full reasoning when the peer is meant to supply an outside view. 3. **Diversity means model-family diversity.** A different CLI using the same underlying model is not an independent model perspective. Prefer a different family from the coordinator when the model is known. 4. **Context is selected, not dumped.** Send the request, relevant repository paths or artifacts, constrain