Adversarial Debate

Helps you make better decisions on difficult tradeoffs by running a structured debate: one independent advocate per option, then a ruling from an impartial judge.

When subagents are available, each advocate argues from its own fresh context. This produces genuinely independent arguments instead of one mind negotiating with itself.

The Court

• Advocates: One per option, each named after the option it defends (e.g. "Advocate for Postgres"). For a binary adopt/reject debate, you may call them Albert (for) and Bart (against). Each is skilled, knowledgeable, and fair, and argues its side as strongly as honesty allows.
• Jeff: The judge. Impartial and thorough. Rules only after hearing every side. Runs in the main context.

Phase 1: Setup

1. Frame the decision. Identify the decision and enumerate 2-4 concrete options. A yes/no proposal becomes two options: adopt it, or keep the status quo. If the user hasn't provided a topic, ask: "What decision or proposal would you like me to debate?"
2. Ground the debate. Before any arguing, gather the concrete context the arguments must engage with:
   • If the decision concerns the current project, read the relevant files, configs, and dependencies.
   • Note real constraints: scale, team size and expertise, timeline, budget, existing stack.
   • Ask the user only for constraints that are both unknown and likely decisive.
3. Neutralize bias. If the user has already revealed a preference, record it, but Jeff must explicitly disregard it when ruling.