# debate-team ## Purpose Structured adversarial debate for contested technical decisions. ## Pattern **Adversarial Debate** — Advocates argue positions, devil's advocate challenges, judge decides. See [docs/PATTERNS.md](../../docs/PATTERNS.md). ## Agents | Agent | Model | Role | Triggers | |-------|-------|------|----------| | debate-advocate-a | sonnet | Use when a contested decision needs structured debate; argues FOR the primary position with evidence and reasoning. | "토론", "A vs B" | | debate-advocate-b | sonnet | Use when debate-advocate-a has stated their position; argues AGAINST or for an alternative position, directly countering advocate-a. | (follows a) | | debate-devils-advocate | sonnet | Use after both advocates have stated positions; challenges the weakest arguments on both sides to prevent groupthink. | (cross-exam) | | debate-judge | sonnet | Use when debate rounds are complete; handles weighing evidence, issuing a verdict, and documenting the dissenting opinion. | (final) | ## Skills | Skill | Trigger | Purpose | |-------|---------|---------| | debate-tradeoff | /debate-tradeoff | Structured debate between two explicit options | | debate-decision | /debate-decision | Open-ended decision — team identifies options | ## Dependencies - **Required:** none ## Routing Rules ### Use this team when: - "트레이드오프", "A vs B", "어느 게 나아", "토론", "반론", "장단점" ### Do NOT use when: - Answer is obvious → respond directly - Research needed first → use research-team first ## For AI Agents ### Entry Point debate-advocate-a receives the question and Option A definition. ### Data Flow advocate-a (R1) → advocate-b (R1, rebuttal) → devils-advocate (challenge) → cross-exam → judge (verdict) ### Exit Criteria debate-judge outputs: decision, reasoning, dissenting view, reversal conditions. Max 5 rounds.