Why one answer is the problem
You ask Claude a question, it gives you an answer that sounds smart, you feel good, you move on. But that answer got shaped by how you asked it — your assumptions, your framing, your emotional lean. Claude picked up on all of it and told you what you wanted to hear.
That's fine for writing emails. It's dangerous for making decisions.
Where this comes from
Andrej Karpathy (co-founder of OpenAI, former head of AI at Tesla) built something called the LLM Council. Instead of asking one model a question, you poll multiple models. Then have them peer-review each other anonymously. Then a chairman synthesizes the final answer.
This skill runs the same method entirely inside Claude Code — using sub-agents with different thinking styles instead of different models. You type "council this"and 5 advisors spin up, argue, review each other's blind spots, and deliver a verdict. All in one session.
The 5 advisors
Not 5 copies of Claude. 5 fundamentally different thinking lenses. They can't converge on the same answer because they're not looking at the same angle.
Why the peer review matters
This is what separates the council from just asking Claude 5 times. After all 5 advisors respond, the skill anonymizes every response — shuffles which advisor maps to which letter. Reviewers don't know who said what. They evaluate on merit.
Then 5 reviewers read all 5 responses and answer 3 questions:
- Which response is strongest and why?
- Which has the biggest blind spot?
- What did all five miss?
That last question is where the gold comes from. Every time the council runs, peer review catches something no individual advisor saw. The gap between perspectives reveals what nobody thought to mention.
The chairman hands you a verdict
One final agent gets everything — all 5 responses, all 5 reviews — and produces the verdict. Not "it depends." Not "consider both sides." A real answer with a concrete first step:
Install it in 10 seconds
Just tell Claude Code to install this skill from the repo below. It'll handle everything.
Tell Claude: "install this skill for me: https://github.com/tenfoldmarc/llm-council-skill"
Prefer to do it manually? Paste this into your terminal:
Then open Claude Code and say "council this: [your question]"
What to council
The council works on any decision where being wrong is expensive and you keep going back and forth.
- Positioning."Should I niche down further or broaden my audience?"
- Pricing."Am I leaving money on the table or pricing myself out?"
- Hiring vs automation."Should I hire a VA or build an automation?"
- Product format."Live workshop or self-paced course?"
- Any fork in the roadwhere you've been circling for days.
If you already know the answer and just want validation — skip it. The council will tell you things you don't want to hear. That's the point.