agent-debate

To coordinate a multi-agent debate, you provide a topic and optional sides to trigger a structured protocol. This generates round-by-round task files and a final synthesis written by Claude to resolve architectural trade-offs.

The debate protocol for multi-agent decision-making enforces explicit pro/con framing across agents like Codex, Gemini, and Claude. It structures topic framing, generates per-round task prompts, and concludes with a bottom-line synthesis.

Yes, you can use Claude, Codex, and Gemini together to surface design decisions. The protocol coordinates these agents to elicit strong pro/con arguments, with Claude generating the final synthesis and storing the transcript.

To store and trace multi-agent debate transcripts, the protocol saves outputs in `.coord/debate_<topic>.md` and `.ai/*.md` task artifacts. This ensures full traceability and reproducibility for your decision-making process.

You should use an adversarial multi-agent debate for policy choices when decisions benefit from explicit pro/con framing. It surfaces clear disagreements and informed syntheses for consequential decisions through structured multi-round arguments.