Make Claude Argue With Itself

• By Praveen Jha · Updated 2026-10-23 · 10 min read · Business & automation

LLM Council - a Claude skill that runs your decision past 5 advisors instead of telling you what you want to hear.

Claude tends to agree with you. This skill fixes that: five advisors analyse your question independently, peer-review each other anonymously, and a chairman delivers one clear verdict - where they agree, where they clash, and what to do first.

5 advisors Anonymous peer review 1 chairman verdict claude.ai + Claude Code Free to use

For the developer handing this on

This guide covers the full LLM Council method, the exact install steps for claude.ai and Claude Code, the trigger phrases, the advisor and reviewer prompt templates, and five copy-paste example prompts. No terminal is needed for the standard install.

THE PROBLEM

One question, one answer - and no way to judge it

You ask one AI a question, you get one answer. It might be great. It might be mid. You cannot tell, because you only saw one perspective.

The council fixes this by splitting the work. Instead of relying on different models, this skill runs inside Claude using sub-agents with different thinking lenses. Each advisor leans fully into one angle, then they review each other, then a chairman synthesises everything.

How a session flows

1. Frame: Your question + workspace context, reframed neutrally.
2. Convene: 5 advisors answer in parallel, 150-300 words each.
3. Review: Answers anonymised A-E; 5 reviewers grade them.
4. Verdict: Chairman synthesises agreement, clashes, blind spots.

What you get back

Where the council agrees - points advisors reached independently. High-confidence signals.

• Where the council clashes - genuine disagreements, presented on both sides, not smoothed over.
• Blind spots the council caught - issues that only surfaced in peer review.
• The recommendation - a real answer, not "it depends". The chairman may side with a lone dissenter if the reasoning is strongest.
• The one thing to do first - a single concrete next step.

Use it where being wrong is expensive. The council shines when there is genuine uncertainty and the cost of a bad call is high. If you already know the answer and want validation, it will likely tell you things you do not want to hear. That is the point.

THE PANEL

The five advisors - thinking styles, not job titles

Each advisor thinks from a fundamentally different angle. They are not personas; they are thinking styles designed to create tension with each other.

1 The Contrarian

Actively looks for what is wrong, missing, or likely to fail. Assumes the idea has a fatal flaw and tries to find it. Not a pessimist - the friend who saves you from a bad deal by asking the questions you are avoiding.

2 The First Principles Thinker

Ignores the surface question and asks "what are we actually trying to solve here?" Strips away assumptions and rebuilds from the ground up. Sometimes the most valuable output is: "you are asking the wrong question entirely."

3 The Expansionist

Looks for upside everyone else is missing - what could be bigger, what adjacent opportunity is hiding, what is being undervalued. Does not care about risk; that is the Contrarian's job.

4 The Outsider

Has zero context about you, your field, or your history. Responds purely to what is in front of them. Catches the curse of knowledge: things obvious to you but confusing to everyone else. The most underrated advisor.

5 The Executor

Only cares whether it can actually be done, and the fastest path to doing it. Ignores theory and strategy. Tests every idea with: "OK, but what do you do Monday morning?" No clear first step, no pass.

Why these five

They create three natural tensions: Contrarian vs Expansionist (downside vs upside), First Principles vs Executor (rethink everything vs just do it), and the Outsider in the middle, keeping everyone honest with fresh eyes.

UNDER THE HOOD

How a council session works - Steps 1 and 2

1 Frame the question (with context enrichment)

Before framing, the skill does two things:

A. Scan the workspace for context (max ~30 seconds). Look for CLAUDE.md, any memory/ folder, files the user referenced, recent council transcripts, and topic-specific files (for a pricing question: revenue data, past launch results, audience research). Pick the 2-3 files that let advisors give specific, grounded advice.

B. Frame the question. Combine the raw question and enriched context into one clear, neutral prompt for all five advisors, covering: (1) the core decision, (2) key context from the message, (3) key context from workspace files, (4) what is at stake. Add no opinion and do not steer it. If the question is too vague ("council this: my business"), ask exactly one clarifying question, then proceed.

2 Convene the council - 5 sub-agents in parallel

Spawn all five advisors simultaneously. Each receives their identity and thinking style, the framed question, and this instruction: respond independently, do not hedge, do not try to be balanced, lean fully into your angle. Target: 150-300 words each.

Copy-ready text

You are [Advisor Name] on an LLM Council.
Your thinking style: [advisor description]
A user has brought this question to the council:
---
[framed question]
---
Respond from your perspective. Be direct and specific.
Don't hedge. The other advisors will cover the angles
you're not covering. 150-300 words. No preamble.

Parallel matters. Sequential spawning wastes time and lets earlier responses bleed into later ones, destroying independence.

UNDER THE HOOD

Steps 3 to 5 - peer review, verdict, delivery

3 Peer review - 5 sub-agents in parallel

This is the step that makes the council more than "ask 5 times". Collect all 5 responses and anonymise them as Response A-E (randomise the mapping to remove positional bias). Each reviewer sees all five and answers, in under 200 words:

1. Which response is the strongest, and why? (pick one)
2. Which response has the biggest blind spot, and what is it?
3. What did ALL responses miss that the council should consider?

Anonymity is non-negotiable: if reviewers know which advisor said what, they defer to certain styles instead of judging on merit.

4 Chairman synthesis

One agent receives everything - the question, all 5 advisor responses (now de-anonymised), and all 5 peer reviews - and produces the verdict in this exact structure:

Copy-ready text

## Where the Council Agrees
## Where the Council Clashes
## Blind Spots the Council Caught
## The Recommendation
## The One Thing to Do First

The chairman must be direct and must not hedge. A clear recommendation is required - and the chairman can disagree with the majority if the dissenter's reasoning is strongest, explaining why.

5 Present the verdict in chat

Present the full verdict directly in the conversation as markdown, headed ## Council Verdict: {short topic}, with the five sections above. Keep it scannable with bullet points. Step 6 (optional): save a transcript only if the user asks, or the decision is significant enough to reference later - as counciltranscript-[timestamp].md in the project's active/ folder.

Worked example - the $297 course. Asked whether to build a $297 Claude Code course for nontechnical solopreneurs, the council agreed the beginner angle has demand but the tool-specific framing fails; clashed on price (too high vs too low); and the peer review surfaced the Outsider's blind-spot catch - "Claude Code" means nothing to the target buyer. Verdict: do not build yet; run a $97 live workshop first, titled by outcome, not by tool.

SETUP

Install it - three routes, no terminal needed

Route A - claude.ai upload (recommended method)

1. Download SKILL.md from github.com/aiwithremy/claude-skills-llm-council - open SKILL.md and click the download button (top right of the file view).
2. Zip it. Create a folder called llm-council, put SKILL.md inside, and compress the folder into llmcouncil.zip.
3. Turn on code execution. On claude.ai go to Settings -> Capabilities and make sure "Code execution and file creation" is on. Skills do not work without it.
4. Go to Customize -> Skills (claude.ai/customize/skills).
5. Upload the skill. Click "+" -> "Create skill" -> "Upload a skill", and choose the ZIP.
6. Toggle it on.

Route B - let Claude install it

Open a new Claude chat and paste:

Copy-ready text

Please install this Claude skill for me. The SKILL.md
file lives in this GitHub repo:
https://github.com/aiwithremy/claude-skills-llm-council
Set it up so I can start using it. Walk me through
anything you need from me.

Claude fetches the file and places it correctly; if your setup needs a manual upload, it will tell you exactly what to click.

Route C - download, then ask Claude

Download SKILL.md as in Route A step 1, then paste: _"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."_

Works in Claude Code and Claude Cowork (claude.ai). Praveen Jha | praveenjha.com

COPY-PASTE LIBRARY

Trigger it - and ask questions worth councilling

Mandatory trigger phrases

council this · run the council · war room this · pressure-test this · stress-test this · debate this

Strong triggers, when paired with a real trade-off: "should I X or Y", "which option", "what would you do", "is this the right move", "validate this", "get multiple perspectives", "I can't decide", "I'm torn between".

Good council questions

• "Should I launch a $97 workshop or a $497 course?"
• "Which of these 3 positioning angles is strongest?"
• "I'm thinking of pivoting from X to Y. Am I crazy?"
• "Here's my landing page copy. What's weak?"

Bad council questions

• "What's the capital of France?" - one right answer.
• "Write me a tweet" - creation, not a decision.
• "Summarise this article" - processing, not judgement.
• Simple yes/no or factual lookups never trigger it.
• "Should I hire a VA or build an automation first?"

Five worked prompts - use them verbatim

Career: Council this: I got a job offer at a startup for 30% more pay, but I'd lose my vesting schedule at my current company. I have 18 months left to fully vest. Should I stay or go?

Business strategy: War room this: We're a 5-person SaaS startup deciding between going upmarket (enterprise sales, longer cycles, bigger deals) vs staying PLG (product-led growth, self-serve, volume). We have 12 months of runway.

Technical: Pressure-test this: I'm building a real-time chat app and deciding between WebSockets vs ServerSent Events. We expect ~10k concurrent users, need message history, and I'm a solo dev.

Life: Run the council: I'm considering moving from Kanpur to Bangalore for better career opportunities in tech, but my family is here and cost of living is significantly higher there.

Product: Stress-test this: Should we launch our MVP with a freemium model or a 14-day free trial? We're B2B, targeting small teams, and our product has a learning curve.

DEVELOPER NOTES

Rules that keep the council honest - and your checklist

Troubleshooting

• Skill never triggers: code execution off (Settings -> Capabilities), skill toggle off, or ZIP missing the llm-council folder.
• Generic answers: framed question lacked context - add stakes, numbers, constraints, workspace files.
• All advisors sound the same: not spawned in parallel, or asked to be balanced. Re-run with the nohedging instruction.
• Verdict says "it depends": re-synthesise - the template requires a clear recommendation and one first step.

Pre-flight checklist

• [ ] SKILL.md downloaded from the GitHub repo
• [ ] Zipped as llm-council.zip with the folder inside
• [ ] Code execution and file creation switched on
• [ ] Skill uploaded under Customize -> Skills and toggled on
• [ ] Tested with one real trade-off decision and a trigger phrase
• [ ] Verdict returned all five sections, ending in one first step

Related pages

• Free AI guides (Resources)
• Learn AI
• Contact Praveen