Skip to content

debate.md

Latest commit

 

History

History
84 lines (76 loc) · 3.28 KB

debate.md

File metadata and controls

84 lines (76 loc) · 3.28 KB
Raw
name debate
description Structured adversarial debate between an advocate and a critic, moderated by a judge. Use for architecture decisions, design reviews, technology choices, or any question where stress-testing a position produces better outcomes than a single perspective.
parameters
name description
topic
The question being debated, e.g. 'Should we migrate from REST to GraphQL for the users API?'
name description
position
The initial position the advocate will defend, e.g. 'Yes, migrate to GraphQL because...'
name description
max_rounds
Maximum debate rounds before the judge must decide. Each round is: advocate → critic → judge.

task: "Debate: {{topic}}" success_criteria: "The judge declares the question resolved with a justified verdict, or max rounds are reached." artifacts:

  • name: debate_log description: "Full debate history — arguments, critiques, and verdicts from all rounds." fields: [role, claims] list: true

steps:

  • type: action name: argue actor: advocate instruction: | The topic: {{topic}} The position you are defending: {{position}}

    Read the debate log for prior rounds. If this is round 1, present your opening argument. If this is a later round, respond to the critic's latest objections and the judge's guidance on what to focus on.

    Write a SHORT structured summary to the debate_log artifact. Keep it under 200 words — key claims only, not the full argument. The full argument goes in your narration text (before the completion tag), not in the artifact. Format: { "role": "advocate", "claims": ["claim 1", "claim 2"] } reads: [debate_log] writes: [debate_log] routes: { done: challenge } max_runs: "{{max_rounds}}"

  • type: action name: challenge actor: critic instruction: | The topic: {{topic}} The position being defended: {{position}}

    Read the debate log. Challenge the advocate's latest argument. Find weaknesses, hidden assumptions, and failure modes.

    Write a SHORT structured summary to the debate_log artifact. Keep it under 200 words — key objections only, not the full critique. The full critique goes in your narration text. Format: { "role": "critic", "objections": ["objection 1", "objection 2"] } reads: [debate_log] writes: [debate_log] routes: { done: evaluate } max_runs: "{{max_rounds}}"

  • type: action name: evaluate actor: judge instruction: | The topic: {{topic}}

    Read the full debate log. Evaluate the latest round of arguments. Decide: is the question resolved, or does it need another round?

    Write a SHORT verdict to the debate_log artifact. Keep it under 100 words. The full reasoning goes in your narration text.

    • If resolved: { "role": "judge", "verdict": "resolved", "conclusion": "one sentence", "prevailing_side": "advocate" or "critic" }
    • If unresolved: { "role": "judge", "verdict": "unresolved", "focus": "what to address next" } reads: [debate_log] writes: [debate_log] routes: resolved: done unresolved: argue max_runs: "{{max_rounds}}"

  • type: terminal name: done outcome: success summary: "Debate concluded with a verdict."