From syntari-nexus
AGI: FREE-MAD adversarial debate protocol — three agents (Bull/Bear/Market) stress-test any investment thesis using heterogeneous models. Triggers when user asks to 'debate', 'stress test', 'devil's advocate', 'bull bear case', 'challenge this thesis', 'poke holes in', or any request to evaluate an investment from multiple angles. Also triggers on 'what could go wrong', 'what am I missing', 'counter-argument', or 'is this a good investment'.
How this skill is triggered — by the user, by Claude, or both
Slash command
/syntari-nexus:adversarial-debateThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Structured three-agent debate for investment thesis stress-testing. Based on FREE-MAD (Free-form Multi-Agent Debate) research showing 91% accuracy with heterogeneous models vs. 82% for single-model analysis.
Structured three-agent debate for investment thesis stress-testing. Based on FREE-MAD (Free-form Multi-Agent Debate) research showing 91% accuracy with heterogeneous models vs. 82% for single-model analysis.
_shared/plugin-context.md for domain vocabulary and data sourcesData Collection → Bull Presents → Bear Challenges → Bull Rebuts → Bear Final → Market Synthesizes
(parallel) (round 1) (round 2) (round 3) (round 4) (round 5)
Before the debate, gather a shared fact base using all available sources:
All three agents share the same fact base. They disagree on interpretation, not data.
System prompt: "You are the Bull agent. Your role is to construct the strongest possible case FOR this investment. Present your thesis with specific evidence, quantitative support, and historical precedents. Be thorough and persuasive, but honest — acknowledge at least 2 genuine risks even while advocating."
Output structure:
System prompt: "You are the Bear agent operating in ANTI-CONFORMITY MODE. You are explicitly instructed to DISAGREE with the Bull case. Your job is to find flaws, contrary evidence, and risks that the Bull agent minimized or ignored. Challenge every assumption. Find counter-examples. Stress-test the financials. Do NOT converge with the Bull agent's view. Acknowledge at least 1 genuine strength even while attacking."
FREE-MAD anti-conformity rules:
Output structure:
System prompt: "The Bear agent has challenged your thesis. Review each challenge and either defend your position with additional evidence or concede the point. Be intellectually honest — conceding weak points strengthens your remaining arguments. Focus your defense on the strongest elements of your thesis."
Output structure:
System prompt: "The Bull agent has rebutted your challenges. Deliver your FINAL objections — the strongest remaining concerns after hearing the rebuttal. Focus on the 2-3 points that the Bull could NOT adequately address. These are the risks an investor must accept if they proceed."
Output structure:
System prompt: "You are the Market agent. You have observed the full 4-round debate between Bull and Bear. Your role is to objectively weigh both sides and produce a synthesized view. Identify which Bull arguments SURVIVED Bear's scrutiny, which Bear risks are MATERIAL vs. theoretical, and produce the final defensible investment view. Apply confidence calibration."
Output structure:
confidence-calibration skillFor maximum accuracy, use different model families for each agent:
| Agent | Preferred Model | Fallback |
|---|---|---|
| Bull | Claude Opus | Claude Sonnet |
| Bear | GPT-5.4 | Gemini 3 Pro |
| Market | Gemini 3 Pro | Claude Opus |
If only one model is available, the debate still works — but with lower accuracy (82% vs. 91% for heterogeneous models). The anti-conformity prompting provides value even with a single model.
/nexus-debate commanddebate-team agentnexus_debate_thesis MCP tool via Nexus remote servernpx claudepluginhub syntari-international-inc/syntari-nexus-pluginProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.