Claude Council
Ask Claude about a decision you've already half-made and it will usually agree with you. This makes five advisors argue about it instead, has them graded, and hands you one verdict.
Free. No email. Works on Claude's paid plans, where Skills live.
Why this exists
Claude's first answer usually agrees with you. Not because it's broken, but because it mirrors the framing you handed it. That's fine for most questions and bad for the ones that matter.
The fix is structural, not a prompt trick. You can't reliably ask one model to "be more critical" — it stays inside your framing. The council makes disagreement a specific agent's job, and no advisor can see what the others said. They can't converge, because they never meet.
1 Install it (about 30 seconds)
Download claude-council.zip above. Don't unzip it.
In Claude, go to Settings → Capabilities → Skills, press upload, and drag the zip in.
Don't see Skills? They're on Claude's paid plans. That's the usual reason it's missing.
2 Use it
Start any chat with "ask the council" and your decision:
ask the council — should I take the senior role at the bigger
company, or stay where I am and ship my own product on the side?It'll ask up to three questions if the decision isn't clear yet, then run. Give it real detail. The messy specifics are where the value is — vague in, generic out.
3 The five advisors
Each one gets the same brief and can't see the others. They're built to pull in different directions:
- 01 · The Contrarian — assumes your idea has a fatal flaw and goes hunting for it. If it genuinely can't find one, it says so, which is its own kind of signal.
- 02 · First Principles — strips the decision down and rebuilds it. Licensed to tell you you're asking the wrong question, and answer the better one.
- 03 · The Expansionist — finds the upside you're overlooking. Every upside has to come with a mechanism, not a vibe.
- 04 · The Outsider — gets your context deliberately stripped out, so it catches the obvious thing you stopped seeing months ago.
- 05 · The Executor — assumes it's decided. First three moves, in order, with timing.
Every advisor also has to write the strongest objection to its own call. That's what stops five confident monologues from reading as five agreements.
4 Then they get graded
Five peer reviewers score every answer on whether it's grounded (follows from what you actually said), useful, and genuinely independent. They flag anything an advisor stated as fact that you never told it.
That last part matters more than it sounds. When I tested this on a real decision, the reviewers caught 34 unsupported assertions across the five advisors — invented percentages, assumed pipelines, a CRM nobody mentioned. Three of the five answers came back only "partly sound."
The system works because the advisors get graded, not because they're individually perfect. The chairman then throws out what didn't survive review, and tells you what it threw out.
5 What you get back
The verdict first, so you can stop reading there if you want:
- The verdict — what to do, and the single most important reason
- What they agreed on
- Where they split, and who wins — resolved, with a side taken. Never "considerations on both sides"
- Discounted — what got thrown out in peer review, and why
- Do this next — concrete moves with rough timing
- What would change this verdict — the one fact that would flip it
Want the argument itself and not just the ruling? Ask for the full advisor transcripts.
One honest heads-up
A full run is eleven agents — five advisors, five reviewers, one chairman. That's a lot more of your usage than a normal chat, and it's genuine overkill for quick questions.
Save it for decisions that actually matter. If you fire it at something small, it'll tell you so and offer to just answer instead.
Built this because my own first answers were too agreeable
I make things like this and show how they work.