/council Open-source Agent Skill for Claude · MIT

Claude Council 80 legendary minds debate your decision. One verdict.

Claude Council is a free, open-source Claude skill by Shagun Choudhary (SuperShary). Warren Buffett, Elon Musk, Steve Jobs, Chanakya and 76 more argue your hard question in character and vote. A High Council hands you one ruling with a 90-day plan.

Install in one line: paste into Claude Code
Install Claude Council by following https://raw.githubusercontent.com/SuperShary/claude-council/main/INSTALL.md

Then start a new session and type /council roster.

Works in Claude Code, Claude.ai, Claude Desktop and the Claude API, and in Cursor, OpenCode, Codex, Gemini CLI and GitHub Copilot through Agent Skills.

A council chamber of 80 seats in 10 coloured councils, all converging on one gold verdict mark THE VERDICT
Claude Council: 80 seats in 10 councils.
  • Money
  • Tech
  • Founders
  • Science
  • Product
  • Growth
  • Strategy
  • Wisdom
  • Leadership
  • Red Team
80 minds · 10 councils · 1 verdict/council <your question>
How it works

One hard question in. One ruling out.

Type /council and your question in Claude Code, or just say “convene the council” anywhere else. Every session follows the same five steps.

  1. Intake

    Your decision in one line: goal, options, constraints and every number you gave. The obvious maths (take-home pay, EMI, runway) is done, and any gaps are listed under Assumed.

  2. Routing

    One council owns the decision, another owns its biggest risk, and deep mode adds the one that owns your hardest constraint. Every council seats a natural opponent.

  3. The councils meet

    Openings, then Crossfire, where members attack each other’s claims by name. Then a vote with conviction scores and the Chair’s brief. No council sees another’s debate.

  4. The High Council

    The Speaker, a neutral moderator, names the fault lines between councils and the Chairs argue them. The Red Team runs a pre-mortem, and the draft ruling is revised.

  5. The Verdict

    One ruling you can act on this week: do X, don’t, or do X only if a measurable condition holds. Plus a plan, tripwires and a vote board.

Quick

Simple choices

A reversible, low-stakes choice in one domain, or “what would X do?”. The Chair rules.

  • 1 council, 3 seats including a natural opponent
  • Openings and a clash of 2–4 lines
  • 350–650 words

Full

Default

Real trade-offs in money, career, product or tech, settled by a High Council.

  • 2–3 councils, 4 seats each, plus a Red Team seat
  • Openings, Crossfire, Vote
  • 1,400–2,400 words

Deep

The summit

Irreversible, life-changing or multi-domain decisions. The Red Team sits as a full council.

  • 3–5 councils, 4–5 seats each
  • Adds a Final word, a pre-mortem and a revised verdict
  • 2,800–4,500 words

In Claude Code, when a subagent tool is available, deep sessions (and full sessions with three or more councils) run each council as its own parallel agent, so no council can anchor on another. Everywhere else, councils run as sealed sessions, one after another. Say quick, full or deep to choose; your word beats the auto-pick.

The councils

Ten councils. Eighty minds.

Each mind has a persona card: its lens, 4–6 documented mental models, the questions it forces you to answer, the red flags it pounces on, who it clashes with, its blind spots and its public sources. Seats are chosen for your question, and at least one natural opponent always sits.

Money

money

Investing, savings, debt, valuation, pricing and fundraising terms.

Chair
Warren Buffett The Oracle
With
Charlie Munger, Ray Dalio, Benjamin Graham, George Soros, Jim Simons, Howard Marks, Rakesh Jhunjhunwala, Cathie Wood

9 minds

Tech

tech

Architecture, AI and ML, infrastructure, scaling, build vs buy, security.

Chair
Elon Musk The Deleter
With
Linus Torvalds, John Carmack, Andrej Karpathy, Jensen Huang, Grace Hopper, Alan Turing, Margaret Hamilton, Werner Vogels

9 minds

Founders

founders

Startups, business models, go-to-market, pivots, quitting a job to build.

Chair
Jeff Bezos The Long-Termist
With
Peter Thiel, Paul Graham, Naval Ravikant, Sam Walton, Ratan Tata, Sara Blakely, Reid Hoffman

8 minds

Science

science

First principles, hard problems, experiments, “is this even possible?”

Chair
Albert Einstein The Thought Experimenter
With
Richard Feynman, Marie Curie, Isaac Newton, Nikola Tesla, Charles Darwin, Srinivasa Ramanujan, Carl Sagan

8 minds

Product

product

What to build or cut, UX, design, focus, craft and launches.

Chair
Steve Jobs The Editor
With
Jony Ive, Dieter Rams, Don Norman, Leonardo da Vinci, Walt Disney, Tony Fadell, Kathy Sierra

8 minds

Growth

growth

Marketing, brand, YouTube and social, sales, offers, pricing psychology, copy.

Chair
David Ogilvy The Adman
With
Seth Godin, Jimmy Donaldson (MrBeast), Alex Hormozi, Gary Vaynerchuk, Robert Cialdini, Dale Carnegie, Estée Lauder

8 minds

Strategy

strategy

Competition, negotiation, positioning, power, timing and leverage in deals.

Chair
Sun Tzu The General
With
Chanakya (Kautilya), Niccolò Machiavelli, Napoleon Bonaparte, Winston Churchill, John Boyd, Carl von Clausewitz

7 minds

Wisdom

wisdom

Career, purpose, fear, regret, burnout, relationships, ethics, discipline.

Chair
Marcus Aurelius The Philosopher-Emperor
With
Seneca, Socrates, Confucius, Lao Tzu, Viktor Frankl, Swami Vivekananda, Maya Angelou

8 minds

Leadership

leadership

Hiring, firing, culture, crisis, feedback and co-founder conflict.

Chair
Satya Nadella The Learn-It-All
With
Abraham Lincoln, Nelson Mandela, Bill Campbell, Indra Nooyi, Andy Grove, A. P. J. Abdul Kalam, Ed Catmull

8 minds

Red Team

red-team

Never picked by topic: the mode seats it. It attacks the draft Verdict for fragility, hidden risk, bias, base rates, fraud signals and anything that could sink you.

Chair
Nassim Nicholas Taleb The Black Swan
With
Daniel Kahneman, Michael Burry, Karl Popper, Jim Chanos, Philip Tetlock, Annie Duke

7 minds

Your own panel/council with Buffett, Musk, Chanakya: <question>
A guest for this session/council add <public figure>
Masks for sharing/council masks <question> swaps names for archetypes like The Oracle.
The Verdict

Anatomy of a Verdict

Full and deep sessions end in the same ten-part Verdict, with the answer shown first; quick sessions get a compact version. “It depends” is only allowed as “X if A, Y if B”, with A and B measurable.

  1. 01
    ⚡ Bottom lineThe ruling in 25 words or fewer, at the very top of the session.
  2. 02
    RulingDo X, don’t, or do X only if Y, with the conditions and the else-branch.
  3. 03
    Why this wonWhose argument carried, and the fault line that lost.
  4. 04
    Your planThis week, the next 30 days, and the day-90 checkpoint.
  5. 05
    TripwiresMeasurable signals with deadlines, each paired with an action.
  6. 06
    Biggest risk → hedgeThe most damaging failure mode and one hedge that caps it.
  7. 07
    Dissent on recordThe strongest opposing member, steelmanned.
  8. 08
    Vote boardOne row per council: █ for, ▒ conditional, ░ against.
  9. 09
    ConfidenceA score out of 10, and the fact or test that would raise it.
  10. 10
    Professional pointerFor money, legal, medical and tax stakes: one line on who should check the plan.
Quick session · Growth council

⚡ Bottom line: Don’t switch to daily Shorts. Fix your titles and first 30 seconds, and test three Shorts a week cut from your long videos.

From the Clash

Godin → Ogilvy: Your split-run Shorts are still ads for strangers. The few hundred who watch everything won’t find you in 30 seconds.

Ogilvy → Godin: Your strangers point is testable, so test it. Fewer than 50 new subscribers in 30 days, and the Shorts go and the two hours return to your regulars.

The Verdict — Ogilvy rules

Ruling: Shorts stay only with 50 new subscribers by day 30; if not → drop them and give the long video those two hours.

Why: MrBeast’s leak argument carried: more uploads can’t fix a 26% watch rate. Ogilvy’s split-run test kept Shorts honest; Godin forced its kill line.

Your plan: This week: find the drop-off second on five videos; retitle your best three. 30 days: fix those intros; apply the kill line. 90 days: judge titles on views per video versus today’s 1,100.

Tripwire: If view duration is under 4:00 by day 30 → pause the Shorts; fix intros only.

Biggest risk → hedge: Shorts pull you toward strangers → cut them only from long videos.

Dissent on record: Godin (Growth, 6/10 against): even three Shorts a week trains you to write for strangers.

Growth  ██░  2 for · 0 conditional · 1 against   avg 7.7

Confidence: 7/10. It rises to 9 once you find the drop-off.

Next moves: “MrBeast, go deeper” · “rematch: my click-through rate is 7%” · “cross-examine Godin vs Ogilvy” · “vote again”

Two Clash lines and the compact Verdict from the skill’s own worked example, lightly trimmed: a creator with 4,200 subscribers asks whether to switch to daily Shorts. Quick mode, so the Chair rules.

Use it

A few commands. Or just ask.

Natural language works everywhere: “use the council”, “convene the council: …” or “what would Munger do?”. Replies come in the language you write in.

Commands

  • /council <question>Picks the mode, councils and seats for you
  • /council quick|full|deep <question>Forces a mode
  • /council money tech <question>Uses exactly the councils you name
  • /council with Buffett, Musk, Chanakya: <question>Your own panel of minds
  • /council masks <question>Archetype names instead of real ones, for shareable output
  • /council rosterLists every council, or one council with /council roster money
  • /council add <public figure>Builds a guest card for this session
  • /council minutesSaves the session to council-minutes/ in Claude Code

Follow-ups

“<name>, go deeper” “rematch: <new facts>” “cross-examine A vs B” “summon <name>” “dismiss <name>” “vote again” “go deep” “unmask”

A rematch re-runs the same seats with your new facts and shows only what moved, plus a Verdict diff. A vote again that flips the majority issues a revised Verdict.

No invented quotes

Every session opens with a one-line simulation notice. The skill tells Claude never to put simulated lines in quotation marks. 70 of the 80 cards carry one verified signature line with its source; the other 10 are paraphrase-only.

Real people stay safe

The rules bar any real person from voicing political endorsements, contested social positions, religious claims, private-life details or allegations, and bar strategy minds from recommending fraud or manipulation.

High stakes still get a ruling

Money, legal, medical and tax questions get a real ruling on your numbers, plus a one-line pointer to the right professional. Crisis signals drop the theatre and point to help.

Install

Install it wherever you use Claude.

Every method installs the same council skill folder. The one-link prompt and the installer scripts never need sudo, and they back up any existing copy first.

One link Paste into Claude Code, or any coding agent with a terminal

macOS / Linux Terminal, no sudo

curl -fsSL https://raw.githubusercontent.com/SuperShary/claude-council/main/install.sh | bash

Windows PowerShell 5.1 or later

irm https://raw.githubusercontent.com/SuperShary/claude-council/main/install.ps1 | iex

Claude Code plugin Inside Claude Code

/plugin marketplace add SuperShary/claude-council
/plugin install council@claude-council

skills CLI skills.sh, requires Node.js

npx skills add SuperShary/claude-council

GitHub CLI v2.90 or later

gh skill install SuperShary/claude-council council --agent claude-code --scope user

Manual Clone and copy the skill folder

git clone https://github.com/SuperShary/claude-council.git
mkdir -p ~/.claude/skills
cp -R claude-council/skills/council ~/.claude/skills/council

Claude.ai and Claude Desktop Pro, Max, Team and Enterprise

  1. Download council.zip from the latest release.
  2. Open Settings → Capabilities and turn on code execution.
  3. Under Skills, choose Add skill → Upload a skill and pick council.zip.
  4. Start a new chat: “Convene the council: …”

Claude API Skills API + code execution tool

# upload once, keep the id
skill = client.skills.create(files=files_from_dir("skills/council"))
# attach on each request
container={"skills": [{"type": "custom",
  "skill_id": skill.id, "version": "latest"}]}
# and enable code execution
tools=[{"type": "code_execution_20250825", "name": "code_execution"}]

The full Python script is in the README →

Other agents Cursor, OpenCode, Codex, Gemini CLI, GitHub Copilot

curl -fsSL https://raw.githubusercontent.com/SuperShary/claude-council/main/install.sh | COUNCIL_ALSO_AGENTS=1 bash

Cursor and OpenCode read ~/.claude/skills, so every method above works for them. Codex, Gemini CLI and GitHub Copilot read ~/.agents/skills: the command above installs to both. Or use the skills CLI with -a codex, -a cursor, -a gemini-cli or -a github-copilot. No /council command? Just ask your agent to “use the council”.

Then start a new session and run /council roster. You should see all 10 councils and 80 minds. Full install guide →

FAQ

Questions, answered.

More in the README.

What is Claude Council?

Claude Council is a free, open-source Agent Skill for Claude, created by Shagun Choudhary (SuperShary). It convenes councils of simulated legendary minds (80 of them in 10 councils, including Warren Buffett, Elon Musk, Steve Jobs, Albert Einstein and Chanakya) who debate your decision in character. A High Council settles their disagreements, a Red Team runs a pre-mortem, and you get one decisive, actionable Verdict.

Who created Claude Council?

Claude Council was created by Shagun Choudhary, known online as SuperShary. He shares his AI builds, Claude skills and automations on his YouTube channel, SuperShary | AI Automation (youtube.com/@SuperSharyG63).

Is Claude Council free?

Yes. Claude Council is free and open source under the MIT license. You only need access to Claude: Claude Code, a Claude plan with Skills (Pro, Max, Team or Enterprise), or the Claude API.

How do I install Claude Council in Claude Code?

Paste Install Claude Council by following https://raw.githubusercontent.com/SuperShary/claude-council/main/INSTALL.md into Claude Code. Or run the installer: curl -fsSL https://raw.githubusercontent.com/SuperShary/claude-council/main/install.sh | bash on macOS or Linux, or irm https://raw.githubusercontent.com/SuperShary/claude-council/main/install.ps1 | iex in Windows PowerShell. It also installs as a Claude Code plugin (/plugin marketplace add SuperShary/claude-council, then /plugin install council@claude-council). Start a new session and type /council roster to check.

Does Claude Council work on Claude.ai, Claude Desktop, the Claude API and other agents?

Yes. On Claude.ai and Claude Desktop (Pro, Max, Team and Enterprise plans), upload council.zip under Settings → Capabilities → Skills with code execution turned on. On the Claude API, upload the skills/council folder with the Skills API and attach it through the container parameter. Cursor and OpenCode read ~/.claude/skills; Codex, Gemini CLI and GitHub Copilot read ~/.agents/skills, so install with COUNCIL_ALSO_AGENTS=1 or npx skills add SuperShary/claude-council -a codex.

Are the quotes in Claude Council real?

The dialogue is simulated: it is reconstructed from each person’s documented public work and is never put in quotation marks. The only real quotes are the verified signature lines on 70 of the 80 persona cards, each used word for word with its source, at most once per session. The other 10 cards are paraphrase-only. Real people are never made to endorse politics, contested social positions or religion.

Is Claude Council financial advice?

No. Claude Council is a decision-thinking tool, not financial, legal, medical or tax advice. It still gives a clear ruling based on your numbers, but it promises no returns and names no stock to buy, and high-stakes Verdicts end with a one-line pointer to the right professional.

How is Claude Council different from asking Claude to role-play experts?

A plain role-play prompt gives you whatever the model improvises. Claude Council adds structure: 80 persona cards built from documented mental models and public sources, a natural opponent in every council, crossfire by name that has to use your numbers, a Red Team pre-mortem, and a fixed Verdict with a plan, tripwires, a hedge, dissent on record, a vote board and a confidence score.

Does Claude Council run code or collect data?

No. The skill is a folder of plain Markdown files, with no scripts, dependencies or telemetry. Claude reads them inside your own session, and your question goes only where your Claude session already sends it. The installers only download the repo and copy the skill folder. In Claude Code, the council writes a file only when you ask: /council minutes, or saving a guest card.

Does Claude Council work in Hindi and other languages?

Yes. Claude Council replies in the language you write in, including the simulation notice, and each persona keeps its own reasoning and cadence. It uses your currency and units, so ₹, lakh and LPA work as naturally as $ and k.

Why doesn’t /council show up after I install Claude Council?

Start a new session, because skills load when a session starts. If you installed the Claude Code plugin, /council works unless another command already uses that name; then use /council:council. For the one-link, installer or manual routes, check that ~/.claude/skills/council/SKILL.md exists and contains name: council. You can always just say “use the council”.

Is Claude Council an MCP server?

No. Claude Council is an Agent Skill: a folder of Markdown files that Claude reads when you call it. There is no server to run, no API key and no code.

The creator

Built by Shagun Choudhary.

Shagun Choudhary (SuperShary)

Shagun Choudhary (SuperShary)

Builds AI automations and Claude skills, and shares his AI builds on YouTube. Claude Council is his open-source project, free for anyone to use, fork and extend.

Need an AI automation or a custom Claude skill? workwithshary@gmail.com

On YouTube

SuperShary | AI Automation

Shagun’s channel for AI builds, agent skills and automations. If Claude Council helped you decide, that channel is where the next builds appear.

youtube.com/@SuperSharyG63