The CouncilThe Claude skill that helps you decide with facts, not opinions.
19 members8 judges vote11 analysts organize facts
What The Council is
A licensed Claude skill. Install it once, type /council and a question, and a panel of 19 named members deliberates on it inside your own Claude.
11 of the 19 are neutral analysts. Their only job is to establish what is true — numbers, sources, precedent — and they never vote. The other 8 are judges, each arguing from one fixed point of view.
A run ends in one verdict — SHIP, PIVOT, or KILL — with a conviction level, the strongest objection against it, the dissent if there was one, and the condition that would flip the answer.
- Brief. The analysts research the question and return facts with sources. No opinions, no recommendation.
- Argue. Every judge reads the same brief and states a position against it.
- Cross-examine. The judges attack each other’s reasoning. Weak arguments are named as weak.
- Verdict. One label, one paragraph of reasoning, and the conditions attached to it.
What a run actually looks like
One real run, replayed. The analysts establish the facts, the judges argue them, and the chairman returns one verdict with the conditions attached.
A scoped run: every judge sat, and 6 of the 11 analysts were called for this question.
Round one — the analysts establish the facts
Competitive Landscape Analystbrief filed
Researcherbrief filed
Market & Customer Insights Analystbrief filed
Financials Analystbrief filed
Channel/Distribution Analystbrief filed
Legal/Compliance Analystbrief filedThe brief
6 analysts researched the question and returned facts before any judge was allowed an opinion. Every judge below argued from this same brief.
Round two — the judges argue the same brief
The Skepticargued
The First-Principles Thinkerargued
The Expansionistargued
The Outsiderargued
The Executorargued
The Customerargued
The CFOarguedbench
The Red TeamarguedbenchWhere they disagreed
“AI writing the email was the expensive part in 2021 and is the free part now.”
“Pricing 20x under the field is not a wedge, it's a signal that the product does 1/20th of the job.”
“If it works, $29 is absurdly underpriced; if it doesn't, $29 is still too much.”
The verdict
PIVOT — the follow-up problem is real, but a $29 email-only standalone can't survive CRM bundling, sub-$60 CAC math, or a market where the money is made in the first five minutes by phone and text.
conviction: medium
Strongest objection
“You are selling the cheapest channel, at the cheapest price, to the most software-fatigued buyer in America — every one of those three choices is individually defensible and together they are fatal.”
Dissent
“Follow-up is the highest-frequency touchpoint an agent has with their database; whoever owns the follow-up thread has a credible path to becoming the CRM, then the transaction layer.”
Kill condition
“20 agents, done-for-you, 30 days, real money on day one. If fewer than 5 convert to paid, the standalone is dead and no amount of product work fixes it.”
Who is in the room
19 specialists. The analysts go and find out what is true and write it into a shared briefing; the judges read that briefing and argue. Analysts never vote. Every run pulls the subset that fits your question — a pricing question does not need the code analyst.
Judges — they argue and they vote.
The two on the bench are called in only when the question needs them.
The Skeptic
judge
Hunts for the strongest real reasons this fails — failure modes, hidden costs, wishful assumptions — and ends with the one objection it would stake its reputation on.
First Principles
judge
Strips the idea back to its underlying mechanics and rebuilds the case from what is actually true, rather than from how things are normally done.
The Expansionist
judge
Looks for the genuine upside others miss — the compounding paths, the 10x version — and says plainly when the honest upside is capped.
The Outsider
judge
Judges on common sense alone, translates every insider term into plain language, and asks the naive questions the experts stopped asking.
The Executor
judge
Asks whether this can actually be done, by you, with these resources, on this timeline — and names the binding constraint and the cheapest test that would settle it.
The Customer
judge
Judges from the buyer's seat: is the problem painful enough to change behaviour, and would I actually pay and switch?
Red Team
judge · bench
Plays your smartest competitor — how they would copy you, undercut you, or out-position you — and names the attack it would actually launch.
The CFO
judge · bench
Interrogates the numbers — margins, CAC against LTV, runway, payback — and names the single number the whole decision hinges on.
Analysts — they report and they never vote.
They establish the shared factual record the judges have to reason from.
Engineer
Explains in plain language what the code does, how it is built, and what state it is really in.
Brand
Describes what is being communicated, to whom, and how the positioning and promise currently read.
Data
States plainly what the numbers say — the trends, the cliffs, the spikes — and flags data too thin to trust.
Competitive
Lays out factually who else operates here, roughly what each offers, and what they charge.
Researcher
Goes and finds what things actually are, quotes the source, cites every URL, and answers the judges' fact requests mid-run.
Market & Customer
Reports the size and direction of demand, and the real voice of the customer from reviews, forums and search.
Financials
Makes the economics legible — revenue, costs, margins, burn — and flags every figure that is assumed rather than known.
Precedent
Finds prior attempts at something meaningfully similar and reports what actually happened to them, successes and failures alike.
Legal
Lays out in plain language the rules that actually apply, and flags clearly where a real lawyer is needed.
UX
Walks the product step by step and describes what using it is genuinely like, including where a user would get stuck.
Channel
Maps the realistic ways this reaches its audience, with rough reach and cost for each.
What you can run
/council <idea>
The full panel
Every member weighs in, then argues. One verdict at the end.
/council facts
Research only
The neutral analysts brief the question. No opinions, no verdict.
/council scoped
Pick your bench
Convene only the members the question actually needs.
20 preset panels ship with your license.
Questions
What is The Council?
The Council is a licensed Claude skill that convenes a panel of nineteen named members to deliberate on a question you bring it. You install it once into Claude, type /council followed by your question, and the panel briefs the facts, argues them, cross-examines its own reasoning, and returns one verdict — ship, pivot, or kill — with the conditions attached to it.
What is the analyst round?
Before any judge is allowed an opinion, eleven neutral analysts research the question and return facts: numbers, sources, precedent, and what is actually knowable. Every judge then argues from that same brief. It means the panel disagrees about what to do rather than about what is true, and it means a verdict can be traced back to something checkable instead of to whoever argued hardest.
Do the analysts vote?
No, and that is deliberate. The eleven analysts establish facts and never cast a vote. The eight judges vote and never research. Separating the two keeps fact-finding from quietly bending toward a preferred conclusion, which is the failure mode when the same voice does both jobs.
What does a verdict actually look like?
One label — ship, pivot, or kill — followed by a paragraph of reasoning, a conviction level, the strongest objection raised against the verdict, the dissent if a judge held one, and the kill condition: the specific thing that, if it turned out to be true, would change the answer. There is a complete real run on the homepage, replayed round by round.
Do I have to convene all nineteen members?
No. /council runs the full panel, /council facts runs the analyst round only and returns a brief with no verdict, and /council scoped convenes just the members a question actually needs. The run replayed on the homepage is a scoped one: every judge sat, and six of the eleven analysts were called.
How long does a run take?
A scoped run usually takes about ten minutes, and a full panel on a genuinely complicated question about thirty. It runs inside your own Claude, so the exact time moves with your plan and with how much the analysts have to research before the judges can start. The comparison that matters is not to a faster answer: it is to the weeks of planning and building that usually stand between a question like this and finding out you were wrong.
Which Claude products does it work with?
Any Claude that supports MCP — Claude Desktop, Claude Code, and Cowork. Installation is one skill file and one license key.
Does it use my own Claude usage?
Yes. The deliberation runs inside your Claude, on your plan, so the depth of a run is bounded by your Claude subscription rather than by us. The skill file is a thin client; the panels, personas, and run procedure live on our server and improve for every licensee at once.
What does it cost, and can I cancel?
$9.99/month. Cancel at any time and your key keeps working until the end of the period you have already paid for, then stops. No annual contract, no seat minimums, and no refunds — cancelling is how you stop paying.
Is my question sent to your server?
No. Your question never leaves your machine. The skill asks our server for three things only — the run procedure, the seat list for a panel, and a persona brief — and each request carries just an identifier, like the panel name or the persona id. There is no field in that interface that could carry your question, and we do not log one. Your license key travels as the authentication token, and we record it to check your subscription is active and to rate-limit abuse. We log only which tool was called, the public licence id, the outcome, a one-way hash of the IP, and the country — never the question, never the answer. See the LICENSE for the full list. The deliberation itself runs entirely inside your own Claude.
Price
$9.99 / month
You are buying a license, not a download. The skill file is a thin client — the panels, personas, and run procedure live on our server and improve for everyone at once. Your key is what connects the two.
Cancel anytime. Your key keeps working until the end of the period you have paid for, then stops. No annual contract, no seat minimums, no lock-in.
What you need
Claude with MCP support — Claude Desktop, Claude Code, or Cowork. The deliberation runs on your own Claude, so the depth of a run is bounded by your Claude plan, not by us.
The Council is licensed software for a single named user. Keys are personal; redistribution or resale is not permitted. Read the LICENSE. Questions: info@goodfellastech.com.
Already a customer? Install & support.