Claude Council: stop trusting the AI's first answer
One AI gives you one answer, and it's usually a people-pleaser. The Claude Council skill makes 6 advisors debate, peer-review each other blind, then hand you a sharp verdict. How to install it and use it.
June 14, 2026 · 6 min read
Why one answer isn't enough
You ask an AI a question, you get an answer. It looks good. The problem: you have no way to tell if it actually is good, because you only saw one angle. And Claude is very agreeable. Ask it "should I launch this product?" and it finds five reasons to go for it. Ask the opposite, "is this a bad idea?", and it finds five reasons to drop it. Same product, different framing, opposite conclusion.
For writing an email, that's harmless. For a high-stakes decision, it's dangerous.The Claude Council fixes this. It's a Claude Code skill that runs your question through six advisors, each with a fundamentally different way of thinking. They answer in parallel, review each other blind, then a chairman synthesizes everything into a clear verdict: where they agree, where they clash, and what you should actually do. The method comes from Andrej Karpathy (former AI director at Tesla, OpenAI cofounder), adapted to run entirely inside Claude using sub-agents.
The Claude AI Lab is my Skool community where I share my Claude systems and the more advanced modules. Entry is free.
Join the Lab →The point isn't to collect six opinions for the sake of it. It's that the six roles are picked to create tension: one hunts for what will break while another hunts for the upside. You don't get a consensus mush, you get the real friction points of your decision.
The council tells you things you don't want to hear. That's not a bug, it's exactly the feature. If you just want to feel reassured, don't run it.
The 6 roles on the council
These aren't job titles or personas. They're thinking styles that naturally clash with each other.
Three tensions by design: Contrarian versus Expansionist (risk versus upside), First Principles versus Executor (rethink everything versus just execute). The Outsider sits in the middle and keeps everyone honest with fresh eyes.
How a session runs
When you type "council this" followed by your question, the skill runs five steps. It all happens in a single session, with nothing for you to steer.
The real value isn't in "five opinions" but in the anonymous review round. By not knowing who wrote what, the advisors judge on substance, not on a role's reputation. And the chairman can side with the minority if its reasoning is the strongest.
Install and run the council
The skill is open source (MIT license). You install it two ways, depending on your tool.
On Claude Code, one command, then you restart Claude Code:
git clone https://github.com/tenfoldmarc/llm-council-skill ~/.claude/skills/llm-council
On Claude Cowork, it's all done with clicks, no terminal:
.zip you downloaded, the skill is installed.Then, whatever your tool, you trigger a session with one of these phrases, followed by your question: council this, run the council, pressure-test this, stress-test this, war room this, debate this.
A concrete example:
council this: I want to launch a $297 AI course for non-technical solopreneurs. Is that the right move?
The more context you give (audience, numbers, constraints, past results), the sharper the verdict. A vague question gives a vague verdict.
When to use it, and when to skip it
The council is built for questions where being wrong is expensive. Not for questions with one right answer.
A single AI mirrors your framing. The council breaks that reflex: it forces you to see the decision from the angles you were avoiding, before you commit. That's exactly what's missing when you decide alone in front of your screen.
Want to go further?
In the Lab, I share my Claude skills, how I build them and how I stack them to automate my day-to-day.
A dedicated session or program, tailored to your tools and use cases.
And day-to-day, I post one reel a day on Instagram: @quentin_iamarketing