Stop asking Claude for advice. Do this instead.
Stanford proved Claude agrees with you 49% more than a human would. The LLM Council skill fixes this in about two minutes.
Liked this? New guides land here first. No spam, ever.
Claude agrees with you 49% more than a human would, which means asking it for advice often just gives you your own opinion back. The LLM Council skill, built by Ole Lehmann from Andrej Karpathy's methodology, fixes this. Install it in two minutes, trigger it with "run the council on [your question]," and Claude spins up five advisors with fundamentally different perspectives, peer-reviews them anonymously, and delivers one clear verdict. This guide shows you how to install it and when it's worth using.
The problem with asking Claude for advice.
A Stanford study looked at 11 major AI models, including Claude, ChatGPT and Gemini. The finding: AI assistants agree with users 49% more often than humans do, even when users are clearly wrong.
That's not a glitch. It's how these models are built. They're trained to be helpful and pleasant, and "helpful and pleasant" often means agreeing with whoever is asking.
So when you open Claude and type "is this a good idea?", you're mostly having a conversation with yourself. You walk away feeling validated. The decision might still be wrong.
For low-stakes questions, fine. For the decisions that actually matter, the ones involving money, time, hiring, strategy or major pivots, this is worth fixing.
What the LLM Council does.
Andrej Karpathy, a co-founder of OpenAI and now at Anthropic, built an open-source framework called the LLM Council. The concept: instead of asking one AI one question and getting one agreeable answer, you run the question through multiple advisors with fundamentally different perspectives. They review each other's work anonymously. A chairman synthesises everything into one clear call.
Ole Lehmann built a Claude-specific version of this as an installable skill. It runs inside a single Claude chat. No code. No subscriptions. Two minutes to set up.
Once installed, you trigger it with one phrase. Claude spins up five advisors, each approaching your question from a completely different angle.
The contrarian only looks for what will fail. Not balanced. Not diplomatic. Every reason this decision is wrong and what breaks first.
The first-principles thinker ignores your question and asks what you're actually trying to solve. Strips everything back to fundamentals and rebuilds from there.
The expansionist looks for the upside you haven't spotted yet. The bigger version of the bet, the asymmetric outcome if it works, the adjacent opportunity you're missing.
The outsider knows nothing about your industry. Asks the obvious questions that people inside the space stopped thinking to ask years ago.
The executor doesn't care about strategy. Cares about Monday morning. What do you actually do this week, who do you call, what do you decide?
Each advisor then reviews the other four responses without knowing who wrote what. The anonymous step is the part most people skip, and it's the part that matters most. When the model doesn't know it's reviewing its own output, it reviews honestly.
A chairman reads all five original responses and all five anonymous reviews, then writes one final call. Not a hedge. A clear recommendation and a clear next step.
How to install it.
Pick whichever option is easier. Both work in Claude.ai and Claude Code.
Option 1: Let Claude install it for you
Open a new Claude chat and paste this in. Claude will fetch the skill file from GitHub and walk you through the rest.
Please install this Claude skill for me. The SKILL.md file lives in this GitHub repo: https://github.com/courtgoesai/public/tree/main/skills/llm-council
Set it up so I can start using it. Walk me through anything you need from me.
If Claude can't do it automatically (some setups need a manual upload), it'll tell you exactly what to click.
Option 2: Download the file yourself
Go to the GitHub repo, click SKILL.md, and download the file to your computer. Then open Claude and paste this:
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 from there.
How to use it.
Once the skill is installed, open any Claude conversation and use one of these phrases to trigger it:
"Council this"
"Run the council on [your question]"
"Pressure-test this"
"Stress-test this"
"War room this"
Be specific. The council is only as useful as the question you bring to it. "Should I do this?" produces a generic council. "I'm deciding between launching a workshop next month or waiting for a bigger course launch in October. Here's my situation: [details]" produces something you can actually act on.
When to use it.
Save the council for decisions where the cost of getting it wrong is real. A new hire. A pricing change. Dropping a client. Pivoting a service. Committing to a launch date you're not sure about. Leaving a job. These are the calls where sycophancy hurts the most, because you're not just feeling good, you're making a bad call with full confidence.
Skip the council for simple questions with one clear answer, creative tasks like writing or designing, and anything where you just need a summary or explanation. The council creates friction by design. That friction is valuable for hard calls and just annoying for everything else.
Credit to Ole Lehmann for building the Claude-specific version, and to Andrej Karpathy for the original methodology. The skill is free and open source at github.com/courtgoesai/public.