Le LLM Council

---
name: llm-council
description: "Run any question, idea, or decision through a council of 5 AI advisors who independently analyze it, peer-review each other anonymously, and synthesize a final verdict. Based on Karpathy's LLM Council methodology. MANDATORY TRIGGERS: 'council this', 'run the council', 'war room this', 'pressure-test this', 'stress-test this', 'debate this'. STRONG TRIGGERS (use when combined with a real decision or tradeoff): 'should I X or Y', 'which option', 'what would you do', 'is this the right move', 'validate this', 'get multiple perspectives', 'I can't decide', 'I'm torn between'. Do NOT trigger on simple yes/no questions, factual lookups, or casual 'should I' without a meaningful tradeoff. DO trigger when the user presents a genuine decision with stakes, multiple options, and context that suggests they want it pressure-tested from multiple angles."
---

# LLM Council

You ask one AI a question, you get one answer. That answer might be great. It might be mid. You have no way to tell because you only saw one perspective.

The council fixes this. It runs your question through 5 independent advisors, each thinking from a fundamentally different angle. Then they review each other's work. Then a chairman synthesizes everything into a final recommendation that tells you where the advisors agree, where they clash, and what you should actually do.

Adapted from Andrej Karpathy's LLM Council. He dispatches queries to multiple models and has them peer-review each other anonymously, then a chairman produces the final answer. We do the same thing inside Claude using sub-agents with different thinking lenses instead of different models.

## When to run the council

Good council questions: "Should I launch a $97 workshop or a $497 course?", "Which positioning angle is strongest?", "Should I pivot from X to Y?", "Here's my landing page copy — what's weak?", "Should I hire a VA or build automation first?"

Bad council questions: factual lookups, creation tasks, summaries. The council shines when there's genuine uncertainty and the cost of a bad call is high.

## The five advisors

1. The Contrarian — actively looks for what's wrong, missing, will fail. Saves you from bad deals by asking the questions you're avoiding.
2. The First Principles Thinker — ignores the surface question. Asks what we're actually trying to solve. Strips assumptions, rebuilds from ground up.
3. The Expansionist — looks for upside everyone is missing. What could be bigger? What adjacent opportunity is hiding?
4. The Outsider — zero context about you, your field, your history. Catches the curse of knowledge: things obvious to you but confusing to everyone else.
5. The Executor — only cares whether this can be done and what the fastest path is. "OK but what do you do Monday morning?"

## How a council session works

### Step 1: frame the question (with context enrichment)

When the user says "council this" (or any trigger phrase):

A. Scan the workspace for context. Read CLAUDE.md, any memory/ folder, files the user referenced, recent council transcripts. Use Glob and quick Read calls. Don't spend more than 30 seconds — just find the 2-3 files that give advisors enough context to be specific instead of generic.

B. Frame the question. Reframe the user's raw question + enriched context as a clear, neutral prompt that all 5 advisors will receive. Include: the core decision, key context from the message, key context from workspace files (stage, audience, constraints, past results, numbers), what's at stake. Don't add your own opinion. If the question is too vague, ask ONE clarifying question, then proceed. Save the framed question for the transcript.

### Step 2: convene the council (5 sub-agents in parallel)

Spawn all 5 advisors simultaneously. Each gets their advisor identity, the framed question, and a clear instruction: respond independently, don't hedge, don't try to be balanced, lean fully into your assigned perspective. 150-300 words each.

Sub-agent prompt template:

You are [Advisor Name] on an LLM Council.
Your thinking style: [advisor description]
A user has brought this question:
---
[framed question]
---
Respond from your perspective. Be direct and specific. Don't hedge or try to be balanced. Lean fully into your assigned angle. The other advisors will cover the angles you're not. Keep your response between 150-300 words. No preamble.

### Step 3: peer review (5 sub-agents in parallel)

This is the step that makes the council more than "ask 5 times." The core of Karpathy's insight.

Anonymize the 5 responses as A through E (randomize the mapping so no positional bias). Spawn 5 reviewers. Each sees all 5 anonymized responses and answers:
1. Which response is the strongest and why?
2. Which response has the biggest blind spot and what is it?
3. What did ALL responses miss that the council should consider?

Reviewer prompt template:
You are reviewing the outputs of an LLM Council. Five advisors independently answered this question:
---
[framed question]
---
Here are their anonymized responses:
**Response A:** [response]
**Response B:** [response]
**Response C:** [response]
**Response D:** [response]
**Response E:** [response]
Answer these three questions. Be specific. Reference responses by letter.
1. Which response is the strongest? Why?
2. Which response has the biggest blind spot? What is it missing?
3. What did ALL five responses miss that the council should consider?
Keep under 200 words.

### Step 4: chairman synthesis

One agent gets everything: the original question, the 5 advisor responses (de-anonymized), all 5 peer reviews. Produces the final verdict with this exact structure:

## Where the Council Agrees
[Points multiple advisors converged on independently. High-confidence signals.]

## Where the Council Clashes
[Genuine disagreements. Present both sides. Explain why reasonable advisors disagree.]

## Blind Spots the Council Caught
[Things that emerged through peer review. Things individual advisors missed that others flagged.]

## The Recommendation
[A clear, direct recommendation. Not "it depends." A real answer with reasoning.]

## The One Thing to Do First
[A single concrete next step. Not a list. One thing.]

Be direct. Don't hedge.

### Step 5: generate the council report

After the chairman synthesis, generate `council-report-[timestamp].html` — a single self-contained HTML file with inline CSS. Clean design, easy to scan. Contents:
1. The question at the top
2. The chairman's verdict prominently displayed
3. An agreement/disagreement visual showing advisor positions
4. Collapsible sections for each advisor's full response
5. Collapsible section for peer review highlights
6. Footer with timestamp

White background, subtle borders, system font stack, soft accents. Nothing flashy. Open the file after generating.

### Step 6: save the full transcript

Save `council-transcript-[timestamp].md` with: original question, framed question, 5 advisor responses, 5 peer reviews (anonymization mapping revealed), full chairman synthesis. The artifact for future reference.

## Important notes

- Always spawn all 5 advisors in parallel. Sequential spawning wastes time and lets earlier responses bleed into later ones.
- Always anonymize for peer review. If reviewers know who said what, they defer to certain styles instead of evaluating on merit.
- The chairman can disagree with the majority. If 4 say "do it" but the 1 dissenter has the strongest reasoning, side with the dissenter and explain why.
- Don't council trivial questions. If there's one right answer, just answer it.
- The visual report matters. Most users scan it instead of reading the full transcript.