Stop trusting Claude's first answer. Run any decision through 5 AI advisors who argue, peer-review each other anonymously, and hand you a verdict you can actually trust.
Based on Andrej Karpathy's LLM Council methodology, adapted to run entirely inside Claude Code using sub-agents with different thinking styles.
Claude is incredibly agreeable. Ask it "should I launch this product?" and it'll find 5 reasons why you should. Ask "is this product a bad idea?" and it'll find 5 reasons why it is. Same product, different framing, opposite answers.
That's fine for writing emails. It's dangerous for making decisions.
When you say "council this", the skill:
- Scans your workspace for relevant context (CLAUDE.md, memory files, etc.)
- Frames your question into a neutral prompt
- Spawns 5 advisors in parallel, each with a different thinking style:
- The Contrarian — hunts for what will fail
- The First Principles Thinker — asks if you're solving the right problem
- The Expansionist — looks for upside you're missing
- The Outsider — responds with zero context (catches curse of knowledge)
- The Executor — only cares what you do Monday morning
- Anonymizes their responses and runs a peer review — advisors review each other without knowing who said what
- Chairman synthesizes the verdict: where the council agrees, where it clashes, blind spots it caught, a clear recommendation, and one concrete next step
- Generates a visual HTML report + full markdown transcript
All in one session. About 4 minutes.
git clone https://github.com/tenfoldmarc/llm-council-skill ~/.claude/skills/llm-councilThen open Claude Code (claude in Terminal).
- Create folder
~/.claude/skills/llm-council/ - Drop
SKILL.mdinside it - Restart Claude Code
Type any of these triggers followed by your question:
council thisrun the councilpressure-test thisstress-test thiswar room thisdebate this
Example:
council this: I'm thinking of pivoting from a $297 course to a $97 live workshop for my audience of non-technical solopreneurs. Is that the right move?
Give it context. The richer the input, the sharper the output.
You'll get:
- A visual HTML report that opens automatically
- A full markdown transcript saved alongside it
Good council questions:
- "Should I launch a $97 workshop or a $497 course?"
- "Which of these 3 positioning angles is strongest?"
- "I'm thinking of pivoting from X to Y. Am I crazy?"
- "Here's my landing page copy. What's weak?"
- "Should I hire a VA or build an automation first?"
Skip the council for:
- Factual questions with one right answer
- Pure creation tasks ("write me a tweet")
- Summaries or processing tasks
- Validation-seeking when you already know the answer
The council tells you things you don't want to hear. That's the feature, not a bug.
- Methodology: Andrej Karpathy's LLM Council
- Adapted for Claude Code sub-agents by @olelehmann
- Published as an installable skill by the community
MIT — do whatever you want with it.