Debate Brainstorm: two-agent structured debate

Take the user's prompt and run it through two independent subagents using the protocol below. Synthesize their final answers into one unified response.

If the user did not provide a prompt, ask what question or plan to debate before starting.

Preserve independence through Round 1. Do not let either agent see the other's work before both original responses are complete. Relay responses and critiques verbatim; do not editorialize them.

Core rules

• Call the subagents Agent A and Agent B.
• Use spawn_agent once for each agent and retain the returned agent IDs.
• Keep both agents read-only during the debate. If the task involves implementation, debate and synthesize the plan first; make changes afterward only if the user requested implementation.
• Run the two calls for each round concurrently when the interface permits.
• Use followup_task to continue with the same agents in Rounds 2 and 3. Never replace an agent mid-protocol merely because its prior turn completed.
• Use wait_agent until both agents have finished each round. Never invent a pending response or advance with only one response.
• If an agent or follow-up fails, retry once. If it fails again, explain which round broke and synthesize the best available material.
• Keep the user updated during waits because the full protocol can take several minutes.