| name | bikeshed |
|---|---|
| description | Run two independent subagents through a structured debate on the same prompt — independent answers, cross-critique, revision, then a synthesized unified response. Use when explicitly invoked as /bikeshed with a prompt, or when the user asks for a two-agent debate/adversarial comparison on a plan, design, or other complex piece of work. Best for plans and complex open-ended work, not simple factual questions. |
Take the user's prompt (everything after /bikeshed), run it through two independent subagents in a debate protocol, and synthesize their final answers into one unified response. If no prompt was given, ask the user what question or plan to bikeshed before starting.
The value comes from independence followed by controlled exchange. Do not let either agent see the other's work before Round 1 is complete, and do not editorialize when relaying material between them — pass responses and critiques verbatim.