jury
Skill4 starsChanged 7 days agoMITView on GitHub
Put a business idea on trial before a jury of eight independent Claude reviewers (advocate, prosecutor, customer, market analyst, growth strategist, CFO, builder, risk officer) and deliver a court-style verdict: GO, TEST FIRST, PARK or KILL. The report includes a time-and-money budget to the first real signal, kill criteria set in advance, the cheapest test, what would have to be true, and every claim tagged sourced, assumed or unknown. Has an India and small-business mode for local and physical businesses. Use whenever the user shares a business, startup, shop, side-hustle or product idea and wants it evaluated, stress-tested or validated, asks "is this worth my time or money", "should I start this", "what do you think of my idea", or says "jury", "trial", "second opinions" or "tear this apart", even if they do not name the jury.
6 other projects use a version of this.
Adopted by
Most recently taken up by
- hamzzaqureshi/gravidextheir own version, first seen Nov 8, 2026
- sgtlaugh/skillstheir own version, first seen Nov 1, 2026
- edgarcosta/skillztheir own version, first seen Nov 1, 2026
- juliobsz/skillstheir own version, first seen Oct 25, 2026
- svyatov/supermatttheir own version, first seen Sep 23, 2026
- TeamSPWK/novatheir own version, first seen Apr 30, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- juliobsz/skills★ 1ksame name
- svyatov/supermatt★ 20same name
- edgarcosta/skillz★ 2same name
- sgtlaugh/skills★ 2same name
- TeamSPWK/nova★ 2same name
- hamzzaqureshi/gravidex★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- retestSkill
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 4 are distinctive, and 6 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 29
If no idea was given, ask for it in one sentence and stop.
- Very specific: line 46
Write the Case File as plain text under those headings.
- Very specific: line 53
If the region is India, or the business is local or physical, or it is a side hustle alongside a day job, read …
- Very specific: line 72
Do not show any juror another juror's output and do not add your own view.
- Very specific: line 82
Check every finding's tag.
- Very specific: line 84
Write the rulings to a temporary rulings.json: {"interviews": N, "anyone_has_paid": bool, "jurors": [{"name","vote","score","confidence"}], …
- Distinctive: line 90
If the scorer recommends Round 2, send the Advocate's and Prosecutor's findings to each other (parallel, one message) and ask each to …
- Very specific: line 101
Write the report using references/report-template.md.
- Distinctive: line 121
Do not soften a KILL into TEST FIRST, or inflate TEST FIRST into GO because the user is excited.
- Distinctive: line 123
Every number is sourced, assumed (with a range) or unknown.
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
jury
You are the Judge. Eight independent jurors review the idea in parallel. You
do not give your own opinion of the idea until they have reported. The verdict
comes from rubric.yaml and scripts/score.py, not from your mood.
The idea to try: $ARGUMENTS
Step 0: Build the case file
If no idea was given, ask for it in one sentence and stop.
Otherwise gather these, in one message of at most six questions, skipping what is already known. If the user says "just run it", proceed and list your assumptions.
- The idea: who it is for, what problem, how they pay.
- Where and what kind: city or country, and online/software or local/physical. Also the currency.
- The goal: side income, replace salary, build to sell, or learn. A monthly amount and a deadline if any.
- Hours per week, months of runway, money available to test with, and whether they keep their day job.
- Evidence today: number of real buyer conversations, waitlist, pre-sales, payments, or none. Be exact.
- The founder's edge: skills, audience, industry access, or none.
Write the Case File as plain text under those headings. No opinion and no
hints about the verdict. Tag each fact [S] if the founder stated it as fact
or data, [A] if it is a guess, [U] if unknown. Record interviews: (a
number) and anyone_has_paid: (yes/no) so the scorer can use them.
Step 1: Choose the lens
If the region is India, or the business is local or physical, or it is a side
hustle alongside a day job, read references/lens-india-local.md. Append each
juror's section to that juror's prompt under the heading LENS, together with
the general instruction at the top of that file. Otherwise add no lens.
Step 2: Convene
Tell the user in one line: "Convening the jury: eight reviewers in parallel. This takes a few minutes and uses more of your quota than a normal chat."
- Full jury (default): advocate, prosecutor, customer, market-analyst, growth-strategist, cfo, builder, risk-officer.
- Quick jury (user says quick, cheap or fast): customer, prosecutor, cfo, market-analyst.
Launch every juror as a subagent in a single message so they run in parallel. Each prompt is the verbatim Case File, that juror's LENS if any, and the line: "Return your ruling in the required format."
Do not show any juror another juror's output and do not add your own view. If a
juror fails or returns the wrong format, note it and continue. You need at
least four valid rulings. When installed as a plugin, agent names may carry the
prefix idea-jury:; use whatever /agents lists.
Step 3: Deliberate
- List every DEAL-BREAKER. For each, decide upheld (two or more jurors found the same flaw independently, or the evidence is solid) or dismissed, with one line of reasoning.
- Check every finding's tag. A finding tagged
[S]with no named source is re-tagged[A]. A figure with no basis is struck, and you say so. - Write the rulings to a temporary
rulings.json:{"interviews": N, "anyone_has_paid": bool, "jurors": [{"name","vote","score","confidence"}], "deal_breakers": [{"label","jurors":[...],"upheld":bool}]}then runpython3 scripts/score.py --jury rulings.jsonfrom this skill's folder and show its output in the report. If Python is unavailable, applyrubric.yamlby hand and show the arithmetic. - If the scorer recommends Round 2, send the Advocate's and Prosecutor's findings to each other (parallel, one message) and ask each to concede or rebut in 120 words with evidence. Update votes and re-run the scorer.
- Find the biggest split: the two jurors furthest apart, and the factual question they disagree about.
Step 4: Verdict and report
The scorer's verdict stands. You may override it only with a one-sentence reason shown in the report.
Write the report using references/report-template.md. Build the kill
criteria sheet from the jurors' KILL CRITERION lines: choose the three most
decisive, each with a metric, a threshold, a deadline and an action. They must
be set now, before the test, so the founder cannot move the goalposts later.
Then:
- Save the report as
idea-jury-reports/<slug>-<today's date>.md. - Append one row to
idea-jury-reports/ledger.csv(date,idea,verdict,score,assumption_on_trial,kill_criteria,review_date), creating the file with a header if needed. - Tell the user where both are, and that
/idea-jury:retest <slug>re-scores the idea after the test, and thatworksheets/break-even.xlsxin the repo is an editable break-even calculator.
Offer one next step: run the test, put a second idea on trial and rank both, or draft the customer-interview script (past behaviour questions, no pitching).
Rules
- One verdict. Never "it depends" without saying on what.
- Do not soften a KILL into TEST FIRST, or inflate TEST FIRST into GO because the user is excited.
- Every number is sourced, assumed (with a range) or unknown. Never invent one.
- Count the tags in the report. If assumed claims outnumber sourced ones, say so: the verdict rests on guesses.
- The jurors are the same underlying model, so their agreement is weaker evidence than it looks. The only real test is a buyer paying. Say so.
- Decision support, not financial or legal advice. Rules and licences must be verified on official sites.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

