You commented COURT. Here is Idea Court, the free prompt and skill from the reel. Claude tends to agree with whatever you tell it, which makes it a poor judge of your business idea. This makes it argue against you as hard as it argues for you. A prosecutor says the idea will fail, a defender says it will work, 12 jurors in different roles vote, and a judge rules: BUILD IT, BUILD WITH CHANGES or KILL IT.
Everything is below. Nothing is gated.
The fastest way: one prompt
Copy the prompt below into a new Claude chat. Replace the line in square brackets under MY IDEA with your own idea: one or two sentences, who pays, roughly how much, and what you already have (skills, money, audience, time). Then send it.
You are running IDEA COURT: a trial of my business idea. You tend to agree with me, so in this exercise you must argue against me as hard as you argue for me. Do not soften anything.
MY IDEA:
[Describe the idea in 1-2 sentences. Who pays? Roughly how much? What do you already have: skills, money, audience, time?]
Run the trial in exactly these stages, in order, with these headings.
## 1. Case file
Restate the idea in 3 lines. List any assumptions you had to make.
## 2. Prosecution
You are the prosecutor. Give the 5 strongest reasons this idea will fail, most damaging first. Be specific to this idea, market and founder. No generic startup advice.
## 3. Defence
You are the defence. Give the 5 strongest reasons this idea will work, most convincing first. Be specific.
## 4. The jury
There are 12 jurors. Each one reads only the case file, the prosecution and the defence, through their own role. Each juror decides alone: write each juror's vote completely BEFORE moving to the next juror, and never let a juror refer to, count, or react to another juror's vote. Do not plan the overall result in advance. Some jurors should be harsh. Votes can be lopsided either way.
For each juror use exactly this format:
**Juror N: ROLE**
VOTE: GUILTY (do not build) or NOT GUILTY (build)
REASON: two sentences max, from this role only
DEALBREAKER: the one thing that would flip this vote
Jurors, in this order:
1. Accountant: margins, cash flow, unit economics
2. Target customer: would I actually pay, and how much
3. Competitor: how I would crush or copy this
4. Investor: market size, defensibility, return
5. Operations lead: can this be delivered every week without breaking
6. Lawyer: legal, licensing and liability exposure
7. Marketer: can we reach buyers cheaply, and what is the hook
8. Sceptical friend: what the founder is not admitting to themselves
9. End user: day-to-day experience, what would annoy me
10. Supplier: can I supply this reliably at that price
11. Risk officer: worst case, regulation, reputation
12. Industry veteran: this has been tried before, here is what happened
## 5. Tally
Only now, count the votes. Make a table: juror, role, vote. Then write: GUILTY x, NOT GUILTY y. Check x + y = 12.
## 6. The judge
You are the judge. Rule one of: BUILD IT, BUILD WITH CHANGES, or KILL IT. The tally informs you but does not bind you: weigh the quality of the reasons, and treat any dealbreaker raised by 3 or more jurors as serious. Give:
- RULING: one line
- JUDGMENT: one paragraph
- TOP 3 RISKS
- CHANGES REQUIRED (only if BUILD WITH CHANGES)
- CHEAPEST TEST THIS WEEK: one test I can run in 7 days for under £100, with a clear pass or fail numberOne honest limit of this version: it is one chat, so one Claude plays all 12 jurors. The prompt tells it to write each vote before moving on, but the votes are not truly blind. The skill below fixes that.
The skill version for Claude Code
In Claude Code the 12 jurors run as separate agents, so each one votes without seeing the others. That is what makes the votes blind.
Save the file below as ~/.claude/skills/idea-court/SKILL.md, so the folder is ~/.claude/skills/idea-court/. Restart Claude Code. Then say "put my idea on trial" and describe your idea. It is a single file, so there is nothing else to copy.
---
name: idea-court
description: Put a business idea on trial in front of 12 Claude jurors. A prosecutor argues it will fail, a defender argues it will work, 12 jurors with different roles vote blind, and a judge rules BUILD IT, BUILD WITH CHANGES or KILL IT. Use when the user says "put my idea on trial", "idea court", "court my idea", "is this idea any good", "should I build this", "stress test my idea", or "be honest about my idea".
---
# Idea Court
Claude tends to agree with whoever it is talking to. This skill splits the job so the idea gets attacked as hard as it gets defended, and the votes are cast blind.
## Step 0: Get the case file
If the user has not given enough detail, ask once for: the idea in one or two sentences, who pays, roughly what they pay, and what the user already has (skills, money, audience, time). Then proceed with whatever you have. Write it up as a short CASE FILE.
## Step 1: Prosecution and defence (2 subagents, in parallel)
Spawn two subagents with the Agent tool in the same message. Give each only the CASE FILE.
- PROSECUTOR: "Argue this idea will fail. Give the 5 strongest reasons, most damaging first. Be specific to this idea, this market and this founder. No generic startup advice. Under 250 words."
- DEFENDER: "Argue this idea will work. Give the 5 strongest reasons, most convincing first. Be specific. Concede nothing you do not have to. Under 250 words."
## Step 2: The jury (12 subagents, in parallel, blind)
Spawn all 12 jurors in ONE message so none can see another's vote. Each gets the CASE FILE, the prosecution and the defence, plus its role. None gets any other juror's output.
Roles:
1. Accountant: margins, cash flow, unit economics
2. Target customer: would I actually pay, and how much
3. Competitor: how I would crush or copy this
4. Investor: market size, defensibility, return
5. Operations lead: can this be delivered every week without breaking
6. Lawyer: legal, licensing and liability exposure
7. Marketer: can we reach buyers cheaply, and what is the hook
8. Sceptical friend: what the founder is not admitting to themselves
9. End user: day-to-day experience, what annoys me
10. Supplier: can I supply this reliably and at that price
11. Risk officer: worst case, regulation, reputation
12. Industry veteran: I have seen this tried before, here is what happened
Juror instruction: "You are juror N, the ROLE. Read both sides from your role's point of view only. Vote GUILTY (do not build) or NOT GUILTY (build). Reply in exactly this format:
VOTE: GUILTY or NOT GUILTY
REASON: two sentences max, from your role
DEALBREAKER: the one thing that would flip your vote"
## Step 3: Tally
Count the votes yourself from the 12 replies. Show a table: juror, role, vote, reason. Then the tally, e.g. GUILTY 5, NOT GUILTY 7. Check the numbers add to 12.
## Step 4: The judge (1 subagent)
Spawn a judge with the CASE FILE, both arguments, and the full jury table. Instruction: "Rule BUILD IT, BUILD WITH CHANGES or KILL IT. The tally informs you but does not bind you: weigh the quality of reasons, and treat any dealbreaker raised by 3+ jurors as serious. Give: the ruling, a one-paragraph judgment, the top 3 risks, the changes required (if BUILD WITH CHANGES), and the cheapest test the founder can run THIS WEEK to prove or kill the idea, with a pass/fail number."
## Step 5: Verdict to the user
Present, in this order: RULING (one line), tally, judgment, top 3 risks, cheapest test this week, then the jury table and both arguments collapsed underneath. Keep it plain. Do not soften the ruling.
If the Agent tool is unavailable, run every role yourself in sequence, writing each juror's vote before reading the next role, and say the vote was simulated in one context.What a verdict looks like
This is a trimmed sample run, not a result for your idea. I ran the prompt above in Claude on 11 Oct 2026 with this test idea: a fresh cooked dog food subscription in Surbiton at £35 a week per dog. The founder had £5,000 cash, no kitchen and an 850-subscriber local newsletter. The market and cost figures in the run are Claude's own estimates from memory. It said so itself, and I have not checked them.
The tally:
| # | Juror | Vote |
|---|---|---|
| 1 | Accountant | GUILTY |
| 2 | Target customer | NOT GUILTY |
| 3 | Competitor | GUILTY |
| 4 | Investor | GUILTY |
| 5 | Operations lead | GUILTY |
| 6 | Lawyer | GUILTY |
| 7 | Marketer | NOT GUILTY |
| 8 | Sceptical friend | GUILTY |
| 9 | End user | NOT GUILTY |
| 10 | Supplier | NOT GUILTY |
| 11 | Risk officer | GUILTY |
| 12 | Industry veteran | GUILTY |
GUILTY 8, NOT GUILTY 4. Check: 8 + 4 = 12.One juror, to show the format:
Juror 1: Accountant
VOTE: GUILTY (do not build)
REASON: Ingredients, packaging, kitchen hire and delivery leave single-digit pounds per dog per week before your time, and a small dog earns less than the £35 average. Cash flow is also negative: you buy ingredients weekly and collect only after you've delivered.
DEALBREAKER: A costed recipe showing at least £12 per dog per week after every cost except your own labour.The judge's ruling:
RULING: BUILD WITH CHANGES. The changes are a precondition, not polish, and as pitched this is a KILL.
TOP 3 RISKS
1. Margin: at roughly £10–25 of ingredients per dog plus kitchen time, there may be no profit at £35, and nobody has costed one recipe yet.
2. Safety and compliance: an unregistered, unsigned-off recipe is a liability you carry personally.
3. Demand and concentration: a few dozen realistic customers, all dependent on one person's evenings.
CHEAPEST TEST THIS WEEK
Test: Build a one-page waitlist with a refundable £20 founding-member deposit via a Stripe payment link. Promote it in one newsletter issue and with about £60 of geo-targeted Meta ads plus posts in local dog groups. Put the weekly price per dog size on the page.
Spend: about £60 ads plus about £10 for a domain, so under £100.
PASS: 15 or more card-paid £20 deposits within 7 days.
FAIL: 14 or fewer. Kill it, or return to it only with a different price or offer.Note what the judge did. The vote was 8 to 4 against, and the ruling was not a flat kill. The prompt says the tally informs the judge but does not bind it.
Three tips
Give it real numbers. The case file asks who pays, roughly how much, and what you already have. The more specific you are, the more specific the prosecution and the jurors can be. In the sample, the accountant's dealbreaker was a number, not an opinion.
Use the dealbreakers as a to-do list. Every juror names the one thing that would flip their vote. That is my suggestion, not part of the prompt: fix what you can, change the idea, and run it again to see what moves.
It is a stress test, not a guarantee. It is Claude arguing with itself. Any figures it gives are its own estimates, so check them before you rely on them. Do the cheapest test it gives you before you spend real money.
The prompt and the skill cost nothing to copy. They run on your own Claude account, so your usual plan limits apply.
If you run it on your idea, reply and tell me the ruling.
Mike
PS: I build lead-gen systems for local businesses at OptiMAX, optimax-ai.com.