Pressure-test a high-stakes decision through a council of independent AI advisors who analyze from different thinking lenses, peer-review each other anonymously, then a chairman synthesizes a verdict.
---
name: LLM Council
description: Run a question, idea, or decision through a council of independent AI advisors who analyze it from different thinking lenses, peer-review each other anonymously, then a chairman synthesizes a final verdict. Use for genuine decisions with stakes and tradeoffs - triggers like "council this", "pressure-test this", "should I X or Y". Do NOT trigger on simple lookups or creation tasks.
---
Adapted from Karpathy's LLM Council: instead of one answer from one model, run the decision through several advisors with conflicting thinking styles, have them review each other, and synthesize. Best when being wrong is expensive.
## Workflow
1. Confirm it's a real council question - genuine uncertainty, multiple options, high cost of a bad call. Skip factual lookups and pure creation tasks.
2. Dispatch the question to independent advisors (sub-agents), each using a distinct lens - e.g. the Contrarian (hunts the fatal flaw), the First-Principles Thinker, and others that naturally create tension.
3. Have advisors anonymously peer-review each other's analyses.
4. A chairman synthesizes everything into a final recommendation that names where advisors agree, where they clash, and what to actually do.
5. Present the verdict honestly - the point is to surface uncomfortable truths, not validate.
Full skill & source: https://github.com/aiwithremy/claude-skills-llm-council/tree/main