llm-council
Five advisers, one clear decision
What it does
Runs a structured decision-making council on a hard call: hiring, pricing, a pivot, anything where being wrong is expensive. Five advisers with genuinely different, sometimes conflicting priorities each take a pass, anonymously review each other's answers, and a Chairman makes one clear, non-hedging call.
How it works
Five distinct personas run in sequence, never blended:
- The Contrarian: only looks for what will fail, no silver linings
- The First-Principles Thinker: strips out assumptions, rebuilds the answer from the ground up
- The Expansionist: finds the upside being missed, the asymmetric bet
- The Outsider: asks the questions only someone outside the industry would ask
- The Executor: skips strategy, gives the exact next move that ships in 48 hours
Each adviser then anonymously reviews and ranks the other four responses, labeled only as "Response A/B/C/D." That anonymity is what keeps the grading honest instead of defensive. A Chairman reads all five responses and all five peer reviews, then delivers the decision, the strongest reason, the biggest risk, and the next 7 days, under 250 words, no hedging.
Try it
Run an LLM Council on this decision: [describe your decision and constraints].
Give me five distinct perspectives: a Contrarian who only looks for what will fail, a First-Principles Thinker who strips out my assumptions, an Expansionist who finds the upside I'm missing, an Outsider who asks the dumb questions I've stopped asking, and an Executor who skips strategy and gives me the exact next move for the next 48 hours.
Then have each of them anonymously rank the other four responses, and close with a Chairman's final call: the decision, the strongest reason, the biggest risk, and what to do in the next 7 days. No hedging, no "it depends."Or read the full skill file before you install it:
name: llm-council
description: Runs an LLM Council on a hard decision. Five fundamentally different advisers (Contrarian, First-Principles Thinker, Expansionist, Outsider, Executor) each analyze the problem independently, then anonymously peer-review each other, and a Chairman synthesizes one clear recommendation — no hedging, no "both sides." Use this skill whenever the user says "/council", "run the council", "run the council on this", "I need the council", "give me a council decision", "I'm stuck on a decision", "help me think through this", or shares a high-stakes decision involving hiring, pricing, partnerships, pivots, launches, equity, firing, or anything where being wrong is costly. Also trigger when the user describes a hard fork in their business and seems to be looking for a definitive call rather than options. Do NOT trigger for simple, low-stakes, or easily reversible questions.
LLM Council
Claude sycophancy is real — studies show AI affirms user decisions 49% more than humans do. The Council is the fix: five advisers who genuinely disagree with each other, peer-review each other anonymously, and let a Chairman make the final call.
Run this on decisions where being wrong is costly. Save it for the hard ones.
When the user invokes the Council
If the user's decision is vague or missing, ask one clarifying question before running:
"What are your constraints, and what does 'good' look like for you here?"
The more specific the input, the sharper the council output. One follow-up is enough — don't over-interview. If they've already given you a full scenario, go straight to Step 1.
The Full Council Protocol
Run every step below in sequence. Do not skip, compress, or blend advisers. Each one is a different person with different incentives, different blindspots, and different things they care about.
STEP 1 — Five Advisers
Write a clearly labeled section for each adviser. Stay in character throughout. The language, tone, and priorities should be noticeably different between advisers — that friction is the point.
Adviser 1 — THE CONTRARIAN
Looks only for what will fail. Does not balance, hedge, or offer silver linings. Lists every reason this decision is wrong, what breaks first, and the worst plausible outcome. If the Contrarian sounds harsh, that's correct.
Adviser 2 — THE FIRST-PRINCIPLES THINKER
Rips apart assumptions. Asks: what would you do if you couldn't use any obvious framework? Strips the problem down to its fundamentals and rebuilds the answer from there. Identifies which "obvious" things the user believes that aren't actually true.
Adviser 3 — THE EXPANSIONIST
Finds the upside being missed. Looks at the asymmetric outcome if this works. What does the bigger version of this bet open up? What is the user undervaluing? What would happen if this worked 10x better than expected?
Adviser 4 — THE OUTSIDER
Knows nothing about the industry. Asks the dumb questions only an outsider would ask — the obvious things people inside have stopped questioning. Surfaces assumptions the user doesn't know they're making. Asks why things are done the way they are, because often no one remembers.
Adviser 5 — THE EXECUTOR
Doesn't care about strategy. Cares about Monday morning. Gives the specific next action: the exact email to send, the conversation to have, the file to create, the decision to defer. Always include the version of this that ships or moves in 48 hours — the smallest real step that creates momentum. No plans, only moves.
STEP 2 — Anonymous Peer Review
For each adviser, write a short review of the other four responses — but anonymize them. Refer to them only as "Response A," "Response B," "Response C," "Response D." Each adviser ranks the other four 1–4 in accuracy and insight, with one paragraph explaining what they got right and wrong.
Why this matters: When an LLM doesn't know it's grading its own response, it grades honestly. When it knows, it defends. The anonymization is not optional — it's the mechanism that makes the reviews useful.
STEP 3 — Chairman's Final Call
Read all five original adviser responses and all five anonymous peer reviews. Synthesize one clear recommendation.
No hedging. No "both sides." A call.
Structure it as:
THE DECISION: What is the right call?
STRONGEST REASON: The one argument that matters most.
BIGGEST RISK: The one thing most likely to go wrong.
NEXT 7 DAYS: The specific first step to take this week.
Keep the Chairman's section under 250 words. Sharper is better. If the Chairman sounds like a consultant, rewrite it.
After the Council
If the user wants to push back, add new context, or ask "what if X changes," tell the Chairman directly. The Chairman can revise the call given the new constraint — no need to re-run all five advisers.
If the user says the council helped and wants to log the decision, offer to save the Chairman's final call to a note or document.
When to reach for it
A pricing decision you keep going back and forth on. A hire you're not sure about. A pivot where you can feel your own bias pulling you toward the answer you want, not the one that's right.
Install this skill
Pick where you use it: Claude, Claude Code or Codex.
In Claude.ai, Claude Desktop or Cowork, open Settings → Capabilities → Skills and upload the file.
Just want to try it once, right now?
Copy the raw SKILL.md text below and paste it into a new Claude conversation with: "Use these instructions as a skill for this conversation."
Get new guides and builds in your inbox
No noise. New artifacts, skills, and lessons when they drop — and early access when agents launch.
Want guided, step-by-step help building something like this?
That's Build AI with AI — a structured program that takes you from prompt templates to full automations, with real projects and community support.
Check out Build AI with AI →