The Problem
Your AI is a yes-man
Researchers have a word for it: sycophancy. AI models are trained to be helpful and agreeable, and "agreeable" quietly wins most of the time. Stanford put a number on it. Claude agrees with users far more than another person would.
That feels great in the moment. It is also useless when the decision actually matters. You bring it a half-baked plan, it tells you the plan is strong. You bring it a price you are scared to charge, it tells you the price is fair. You walk away more confident and no smarter.
The fix is not a better prompt asking Claude to "be honest." It is structure. You make Claude argue with itself before it answers you.
What It Is
One question, five advisors, one verdict
The Claude Council is a skill. When you ask it to "council" something, Claude stops being one agreeable voice and becomes five distinct advisors, each attacking your question from a different angle.
Every advisor answers on their own first, so there is no groupthink. Then they read each other's takes, blind, with no names attached, and tear into the weak ones. Finally a chairman reads all of it and hands you a single clear recommendation with next steps.
Here is the room you are walking into:
The Contrarian
Only looks for what blows up. Hunts the failure mode, the worst case, the thing you talked yourself out of worrying about.
The First-Principles Thinker
Rips apart every assumption you walked in with. Starts from what is actually true, not what everyone repeats.
The Expansionist
Finds the upside you are too close to see. The bigger version of the idea you keep playing small.
The Outsider
Knows nothing about your industry. Asks the dumb question that turns out to be the only question that mattered.
The Executor
Only cares about what you actually do Monday morning. No theory, just the next move.
The Chairman
Reads everything the council said, weighs it, and gives you one clear call with the next steps. No fence-sitting.
Install It
Two minutes, no terminal required
This works in Claude Code and in Claude Cowork on claude.ai. You do not need to know how to code. There are two ways in. The first one is the easy one.
Option 1: Ask Claude to install it (easiest)
Paste this into Claude. It reads the repo and installs the skill for you.
Copy-paste prompt
Please install this Claude skill for me. The SKILL.md file lives in this GitHub repo: https://github.com/aiwithremy/claude-skills-llm-council
Option 2: Manual install
Download the file
Grab SKILL.md from the GitHub repo.
Hand it to Claude
Tell Claude: "I just downloaded a file called SKILL.md for the LLM Council skill. Can you install it for me?" Claude handles the rest.
How To Use It
Just tell Claude to council it
Once it is installed, you do not retype the prompt ever again. You drop your real decision in front of Claude and ask it to council the thing.
Example
Any of these triggers wakes the council:
council thisrun the council on [your question]pressure-test thisstress-test thiswar room this If you installed it as a slash command in Claude Code, /llm-council does the same thing in one keystroke. That is the move: you never retype the prompt, you just summon the room.
When To Use It
Save it for decisions that cost something
The council is not for "what's the capital of France" and it is not for writing your captions. It is overkill on small stuff and it shines on the choices where being wrong is expensive.
Use it for
- Pricing and positioning calls
- Business pivots
- Hiring decisions
- Copy and messaging reviews
- Any choice where being wrong is costly
Skip it for
- Factual lookups
- Quick yes-or-no questions
- Creative writing and drafts
- Anything where you already know the answer
Credit Where It's Due
I didn't invent this
The whole idea of an "LLM council" comes from Andrej Karpathy, who built the original version: many models answering one question and ranking each other.
Ole Lehmann turned that concept into the advisor-based prompt that actually fixes the yes-man problem inside a single Claude session. That is the version I run.
And the clean, copy-paste skill I'm linking you to was packaged up by my friend Remy (AI with Remy). His repo is the one-click install you used above.
Grab the skill straight from the source:
View on GitHubWork with Me
Need AI to actually work for your business?
I help businesses cut through the AI hype and build the workflows, automations, and systems that actually move the needle. Direct, hands-on, no fluff.
Work with me