debatelisted

# Debate Runs a bounded, structured multi-agent debate on a contested decision and returns a single recommended answer (or an honest "no clear winner" when positions are genuinely tied). This skill is self-contained and uses Claude Code's native subagent primitives — no external team-manager dependency. ## When to use this vs `superpowers:brainstorming` | Use `debate` when | Use `brainstorming` when | |---|---| | Positions are nameable upfront (A vs B, or 2–4 named options) | Analysis is open-ended; positions emerge from exploration | | You want a decision | You want a trade-off map and recommendation from parallel lenses | | The question is contested (multiple defensible answers exist) | You need Explorer/Analyst/Critic working in parallel without messaging | If the topic is open-ended design ("design our notification system") or has no real contest ("should we use version control"), stop and suggest the alternative. ## Team Composition **Fixed at 4 agents** for binary and 3-position topics: | Slot | Role | Spawn mechanism | |---|---|---| | Debater A | Pinned to Position A for the entire run | `Agent(subagent_type="general-purpose", name="debater-a", run_in_background=true, ...)` | | Debater B | Pinned to Position B for the entire run | `Agent(subagent_type="general-purpose", name="debater-b", run_in_background=true, ...)` | | Debater C | Pragmatist (binary) OR Position C (3-option) | `Agent(subagent_type="general-purpose", name="debater-c", run_in_background=true, .