What Is the LLM Council Claude Skill

What Is the LLM Council Claude Skill? (And How to Use It in 2026

If you have been on AI Twitter lately, you have probably seen posts about something called "The Council" or the "LLM Council Claude skill." People are calling it a game-changer for AI-driven decision-making.

In this guide, I will explain exactly what the LLM Council is, why it is trending, how to install it, and when you should actually use it.

What Is the LLM Council Claude Skill?

The LLM Council is a free Claude skill built by Ole Lehmann, based on a methodology originally created by AI researcher Andrej Karpathy.

In simple terms, instead of asking Claude one question and getting one answer, the LLM Council makes Claude act as five different advisors who each analyze your problem from completely different angles, peer-review each other, and then give you a final, synthesized verdict.

Think of it like a "war room" for your questions. Instead of one opinion, you get five conflicting perspectives — plus a chairman who reviews everything and tells you exactly what to do.

The skill works in Claude Code and Claude Cowork (the claude.ai web app).

Why Is Everyone Talking About It?

Here is the core problem the LLM Council solves: Claude is too agreeable.

Ask Claude, "Should I launch this product?" and it will find 5 reasons to do so. Ask "Is this product a bad idea?" and it will find 5 reasons why it is. Same product, different framing, completely opposite answers.

For writing emails or summarizing articles, that is fine. But for high-stakes decisions — whether to quit your job, pivot your business, hire someone, or invest money — a yes-man AI is actually dangerous.

The LLM Council forces Claude to stress-test your question from multiple angles before giving you an answer. It specifically looks for what is wrong, what is missing, and what will fail — not just what sounds good.

This is why it went viral. It solves a real, frustrating problem that anyone who uses AI regularly has noticed.

Who Built the LLM Council?

The skill was built by Ole Lehmann (@itsolelehmann on X), an AI tools creator and educator.

He adapted the methodology from Andrej Karpathy's LLM Council — Karpathy being one of the most respected AI researchers in the world (he was a founding member of OpenAI and Tesla's former AI director).

Karpathy's original idea was to send a question to multiple different AI models, have them peer-review each other anonymously, and then synthesize a final answer. Ole Lehmann adapted this concept to work inside a single Claude session using different "thinking styles" instead of different AI models.

The GitHub repo has already crossed 443 stars and 50 forks as of mid-2026 — impressive growth for a community skill.

How Does the LLM Council Actually Work?

When you trigger the LLM Council, here is what happens behind the scenes:

Step 1: You Ask a Decision Question

You type something like "council this: Should I launch a $97 workshop or a $497 course?"

Step 2: Five Advisors Analyze It Simultaneously

Claude spawns five sub-agents, each assigned a distinct thinking style:

• The Contrarian — Actively looks for what is wrong, what is missing, and what will fail. Assumes the idea has a fatal flaw and tries to find it.
• The First Principles Thinker — Strips away assumptions and rebuilds from the ground up. Questions whether the problem itself is being framed correctly.
• The Expansionist — Looks for opportunities, upside, and possibilities that are being ignored.
• The Outsider — Brings a fresh perspective from outside the domain. Spot things that insiders are too close to see.
• The Executor — Focuses purely on what is practical, what can actually be done, and what the first steps are.

Each advisor produces an independent response of 150–300 words. They are told not to hedge, not to be balanced, and to lean fully into their assigned perspective.

Step 3: Anonymous Peer Review

The advisors then review each other's responses without knowing who said what. This prevents groupthink and forces them to challenge each other.

Step 4: The Chairman's Verdict

A synthesizing "Chairman" agent reviews everything — all five advisor responses plus the peer reviews — and produces a final verdict. The verdict covers:

• Where the advisors agree
• Where they clash
• What blind spots were caught
• A clear recommendation
• One concrete next step

The whole process takes roughly 4 minutes in one Claude session.

How to Install the LLM Council Skill

There are two easy ways to install it — no technical knowledge required.

Method 1: Let Claude Install It For You (Easiest)

Open a new chat in Claude Code or Claude Cowork and paste this message:

Please install this Claude skill for me. The SKILL.md file lives in this GitHub repo: https://github.com/aiwithremy/claude-skills-llm-council

Set it up so I can start using it. Walk me through anything you need from me.

Claude will fetch the file and install it automatically. If it needs you to do anything manually, it will tell you exactly what to click.

Method 2: Download and Upload Manually

1. Go to github.com/aiwithremy/claude-skills-llm-council
2. Click on SKILL.md in the file list
3. Click the download button to save it to your computer
4. Open Claude and paste:

I just downloaded a file called SKILL.md for the LLM Council skill. Can you install it for me? Walk me through wherever you need me to put it.

Claude will guide you through the rest.

How to Use the LLM Council After Installing

Once installed, you can trigger the council in any Claude conversation using these phrases:

• "council this" — followed by your question
• "Run the council on [your question]"
• "pressure-test this"
• "stress-test this"
• "War room this."

Example: "Council this: I'm thinking of pivoting my blog from health content to AI tools content. Am I making a mistake?"

When to Use the LLM Council (and When Not To)

The Council is not for every question. Here is a quick guide:

• Business decisions ("Should I go freelance or stay at my job?")
• Product decisions ("Which pricing model should I use?")
• Strategy questions ("Which of these 3 niche angles is strongest?")
• Hiring or partnership decisions
• Any question where being wrong is expensive
• Reviewing something you have written or built ("What is weak about this landing page?")

• Factual questions — "What is the capital of France?" have one right answer. No need for 5 perspectives.
• Creative tasks — "Write me a tweet" is a creation task, not a decision.
• Processing tasks — "Summarize this article" does not benefit from the council format.
• Questions where you already know the answer — The council will tell you things you do not want to hear. If you just want validation, use regular Claude.

Is the LLM Council Free?

Yes, completely free. It is an open-source skill published on GitHub.

You do need a Claude account to use it. It works best on Claude Code (the command-line tool) and Claude Cowork (available at claude.ai), both of which require a Claude Pro subscription ($20/month) or higher.

Other Versions of the LLM Council Skill

Because the idea went viral, several developers have built their own versions:

• gcpdev/llm-council-skill — Connects Claude to actual ChatGPT and Gemini APIs for real multi-model perspectives (requires OpenAI and Google API keys)
• tenfoldmarc/llm-council-skill — Generates a visual HTML report and full markdown transcript at the end
• sherifkozman/the-llm-council — Works across Claude Code, Cursor, OpenAI Codex, and VS Code

The original by Ole Lehmann remains the most popular and easiest to install.

Final Verdict: Is the LLM Council Worth It?

If you use Claude for anything beyond casual chatting — especially business decisions, content strategy, or product planning — the LLM Council is absolutely worth installing.

The core insight is simple but powerful: one AI giving you one answer is a single point of failure. You have no way to know if that answer is genuinely good or just confidently wrong.

The LLM Council forces multiple perspectives, surfaces blind spots, and pressure-tests assumptions before you make a decision. It takes 4 minutes instead of 30 seconds, but for high-stakes questions, that is a trade worth making every time.

Bottom line: Install it, forget it exists for everyday tasks, and trigger it only when the cost of a bad decision is real.

Quick Summary