Run a structured multi-agent debate. Spawns debaters for each position, runs opening arguments and rebuttals, then a 3-judge panel picks the winner.
Parse the user's input into a topic and 2+ positions. If fewer than 2 positions, ask before proceeding.
Spawn one debater subagent per position, all in parallel. Give each the topic, their assigned position, and tell them this is the opening round — no rebuttals yet.
Spawn one debater subagent per position again, all in parallel. This time include the other positions' Round 1 arguments so each debater can rebut.
Spawn 3 judge subagents in parallel. Each gets the same prompt: the topic and the complete, unabridged output from every debater across both rounds. Do not summarize, truncate, or paraphrase the arguments. Judges must not see each other's output.
Synthesize the judges' rankings yourself — do not delegate this step.
Use a Borda count: 1st = N points, 2nd = N-1, etc. Sum across judges.
Present a scorecard table, declare the winner, note whether the decision was unanimous or split, and summarize the key reasoning across judges.