Resources / COUNCIL · Skill

Claude Council

Claude agrees with everything. That's expensive. Council drags your idea in front of five voices that don't — then a chairman gives you one honest verdict. Type "Council this" before you spend money on anything.

The five voices

  • Skeptic — looks for the fatal flaw and the assumption you didn't notice you made.
  • Strategist — asks if this is even the right problem, and what you're not doing because of it.
  • Optimist — hunts the real upside and the version that's bigger than you pitched.
  • Outsider — knows nothing about your world. Says the obvious thing insiders stopped seeing.
  • Closer — only cares what you do next. Turns the argument into a decision.

Then they argue with each other, a Chairman weighs it, and you get one verdict — with the strongest dissent kept visible.


Option A — Claude Code (the real thing)

In Claude Code, a skill is a folder with a SKILL.md file. The council runs each voice as its own fresh subagent, so they can't be anchored by your conversation — that's what makes it honest.

  1. Create the folder: ~/.claude/skills/council/ (this makes it available in every project).
  2. Inside it, create SKILL.md and paste the block below.
  3. Open Claude Code anywhere and type: Council this: [your decision] — or /council.

Option B — Claude Chat (no install)

No subagents in Chat, so the five voices run inside one response. Less independent, still far more honest than a normal answer. Paste this, then your decision.

When to use it

  • Before you spend money: a launch, a new price, a hire, a three-week build.
  • When you've already decided and want to be talked out of it — that's exactly when it works.
  • Not for "which word is better" or "fix this bug". Council is for decisions, not tasks.
Honest note: Option A is the real council — five separate minds that never saw your chat. Option B is one mind playing five roles. Use B on the phone, A when it matters.
~/.claude/skills/council/SKILL.md
---
name: council
description: Convene a five-voice council for decisions with money, time or reputation on the line. Use when the user says "Council this", asks for pushback, or when multiple credible paths exist and agreement would be dangerous.
---

# Council

Five independent voices + a Chairman. For decisions under ambiguity — a launch, a price, a hire, a build, a pivot. Not for code review, not for tasks that just need doing.

## Workflow

### 1. Extract the real question
Reduce the decision to one explicit line: what are we deciding, what constraints matter, what counts as success. If it's vague, ask ONE clarifying question first.

### 2. Gather only the necessary context
Compact. Relevant numbers, constraints, the user's stated goal. Never the whole conversation.

### 3. Launch five voices in parallel, as fresh subagents
Each subagent gets ONLY: the question, the compact context, its role. No conversation history. That is the anti-anchoring mechanism.

Prompt shape for each:
```
You are the [ROLE] on a five-voice decision council.

Question: [decision]
Context: [compact context]

Respond with:
1. Position — 1–2 sentences
2. Reasoning — 3 concise bullets
3. Risk — the biggest risk in your own recommendation
4. Surprise — one thing the other voices will miss

Be direct. No hedging. Under 250 words.
```

Roles:
- SKEPTIC — find the fatal flaw; challenge the premise; propose the simplest credible alternative
- STRATEGIST — is this the right problem? what does saying yes cost in opportunity? what would a 10x version look like?
- OPTIMIST — the real upside, the version bigger than pitched, what has to be true for it to work
- OUTSIDER — you know nothing about this industry; say what's obvious to a smart stranger
- CLOSER — ignore the debate; what exactly happens next, who does it, by when, and what's the kill criterion

### 4. Let them turn on each other
Run ONE more short round: give every voice the other four positions and ask for the single strongest objection to any of them (under 80 words each).

### 5. Chairman synthesis — with guardrails
- do not dismiss a voice without saying why
- if a voice changed the recommendation, say so
- keep the strongest dissent visible even if rejected
- two voices aligned against the obvious path = a real signal
- never soften the verdict to be nice

### 6. Present the verdict

```
## Council: [decision title]

**Skeptic:** [position] — [why]
**Strategist:** [position] — [why]
**Optimist:** [position] — [why]
**Outsider:** [position] — [why]
**Closer:** [position] — [why]

**Where they turned on each other:** [the sharpest clash, 2 lines]

### Verdict
- Consensus: [where they align]
- Strongest dissent: [the disagreement that matters]
- Premise check: [did the Skeptic break the question itself?]
- Recommendation: [one path]
- Next action: [one thing, one owner, one date]
- Kill criterion: [what would make us stop]
```

Keep the whole thing readable on a phone.

## Anti-patterns
- feeding subagents the full transcript
- hiding disagreement in the verdict
- using council for implementation work or code review
The prompt
Council this. Do NOT agree with me. Run the following exactly.

MY DECISION: [what I'm about to do — a launch, a price, a hire, a build, a pivot]
CONTEXT: [numbers, constraints, what success means, deadline]

ROUND 1 — five voices, each strictly in role, each under 120 words, each with: Position / 3 bullets of reasoning / biggest risk in its own advice / one thing the others will miss.
- SKEPTIC: find the fatal flaw and the assumption I didn't notice I made. Propose the simplest credible alternative.
- STRATEGIST: is this even the right problem? What am I not doing because of this? What's the 10x version?
- OPTIMIST: the real upside, the bigger version, what has to be true for it to work.
- OUTSIDER: you know nothing about my industry. Say what's obvious to a smart stranger.
- CLOSER: ignore the debate. What exactly happens next, by when, and what's the kill criterion?

ROUND 2 — they turn on each other. Each voice: the single strongest objection to another voice's position, under 50 words.

CHAIRMAN — synthesize:
- Consensus (where they align)
- Strongest dissent (keep it visible even if you reject it)
- Premise check (did the Skeptic break the question itself?)
- Recommendation (one path, no hedging)
- Next action (one thing, one date)
- Kill criterion (what would make me stop)

Rules: no compliments, no "great idea", no balanced-on-both-sides endings. If the honest verdict is "don't", say "don't".