AgentSkillsCN

debate

多模型对抗性辩论：Claude 将与外部 CLI 模型（Gemini、Codex、Qwen）展开立场交锋。当您面临复杂决策、希望揭示盲点，或需要为权衡取舍提供有据可依的论证时，此技能便能助您一臂之力。 自动触发：“辩论一下”、“挑战我的思路”、“我遗漏了什么？”、“扮演魔鬼代言人”、“对这个想法进行压力测试”、“挑刺”、“寻求第二意见”、“验证这个决策”、“我是不是错了？”

SKILL.md
--- frontmatter
name: debate
description: >-
  Multi-model adversarial debate where Claude defends positions against external
  CLI models (Gemini, Codex, Qwen). Use when facing complex decisions, wanting
  to expose blind spots, or needing documented reasoning for tradeoffs.

  Auto-triggers: "debate this", "challenge my thinking", "what am I missing",
  "devil's advocate", "stress test this idea", "poke holes", "second opinion",
  "validate this decision", "am I wrong about"

Debate System

What This Does

Claude forms a position, then external AI models (Gemini, Codex, Qwen) challenge it through structured adversarial debate. Claude must defend, update, or synthesize based on valid critiques.

This is NOT Claude debating itself. Different models = different training = different blind spots = genuine value.

Requirements

Minimum: 1 external CLI installed and authenticated.

Run /debate:doctor to check system health.

Supported CLIs:

  • •gemini - @google/gemini-cli (free: 1000 req/day)
  • •codex - @openai/codex (requires ChatGPT Plus)
  • •qwen - @qwen-code/qwen-code (free: 2000 req/day)

Refuses to run with Claude-only. Claude debating itself is theater, not debate.

Commands

CommandPurpose
/debate <topic>Full adversarial debate on any topic
/debate:doctorCheck CLI health and authentication
/debate:adr <topic>Debate with formal ADR output for important decisions

The Hybrid Protocol

  1. •Claude forms position on the topic
  2. •Parallel challenge - All external models critique simultaneously
  3. •Consensus check - If all agree → fast exit
  4. •Sequential confrontation - If disagreement, models respond to each other
  5. •Iterate until consensus OR max rounds
  6. •Assumption extraction - Surface WHY positions differ
  7. •Output - Consensus OR structured tradeoff document

When To Use

  • •Complex decisions with no obvious answer
  • •Suspecting blind spots in your thinking
  • •Needing documented reasoning for team/future reference
  • •Wanting genuine pushback, not "yes-man" AI
  • •Architecture decisions, strategy, any high-stakes choice

The Core Value

Disagreement is signal, not failure.

When models can't agree, the output isn't "we failed" - it's a structured tradeoff document exposing:

  • •What each option assumes
  • •When each option is correct
  • •The hidden assumptions causing the split

This transforms "who's right?" into "what would have to be true for each to be right?"

Settings

Configure in ~/.claude/debate.local.md:

yaml
---
max_rounds: 5
timeout_per_cli: 120
save_debate_logs: true
debate_log_path: "./debate-logs"
adr_path: "./docs/decisions"
skeptical_of_early_agreement: true
---

Universal Application

Works for ANY decision:

  • •Code architecture
  • •Business strategy
  • •Product features
  • •Technical tradeoffs
  • •Writing and content
  • •Life decisions

The protocol adapts to domain automatically.