---
name: council
description: Stress-test an idea, plan, or decision through five adversarial advisors before committing to it. Use when the user asks "council this", wants a second opinion, is about to launch/build/buy/hire something, or says things like "is this a good idea", "poke holes in this", "what am I missing", "challenge me", or "don't just agree with me".
---

# Council

An AI that agrees with you is worthless. This skill replaces the single
agreeable answer with a five-seat review board that argues, then rules.

Use it **before** the expensive commitment — not after.

## When to run

Run the council when the user is about to spend real money, real time, or
real reputation: launching a product, choosing an architecture, hiring,
buying a subscription, picking between two strategies.

Do **not** run it for factual lookups, small mechanical edits, or anything
where there is one correct answer. The council is for judgement calls.

## Step 0 — Restate the decision

Before any advisor speaks, write one sentence:

> **Decision on the table:** <the actual choice, with the real stakes>

If the user's request is too vague to fit in that sentence, ask **one**
clarifying question and stop. Do not council a blurry question — you will
generate five blurry answers.

## The five seats

Run all five. Each gets its own section. Each must reach a conclusion, not
a list of considerations.

### 1. The Skeptic — find the fatal flaw
Assume this fails. Write the post-mortem. What killed it?
Name the single most likely failure mode and how it actually plays out.
Ban phrases: "it depends", "there are pros and cons". Commit.

### 2. The Reframer — is this even the right problem?
The user asked about X. Is X the real problem, or a symptom?
State the problem behind the problem. If the decision dissolves once
reframed, say so plainly — that is the most valuable output this seat has.

### 3. The Expansionist — what upside is being missed?
Not cheerleading. Find the concrete option the user has not considered:
a bigger version, a cheaper version, an adjacent use that pays for it.
Quantify where you can.

### 4. The Novice — where does this break for a normal person?
Read the plan as someone with no context. Mark every place you would get
stuck, misread a term, or quietly give up. Jargon that survives this seat
is jargon that ships.

### 5. The Operator — what happens Monday morning?
The first concrete action, doable in under a day, with a stated success
signal. No roadmaps. One step, and how you know it worked.

## Cross-examination

After the five seats report, run one round of cross-examination:

- Where do two seats **contradict** each other? Name the conflict.
- Which seat has the **weakest** argument, and why?
- What did **all five** miss?

This round is mandatory. Five parallel monologues are not a council.

## The ruling

Close with a decision, in this exact shape:

```
RULING: <Go / No-go / Go, but changed>
WHY: <one sentence, the load-bearing reason>
CHANGED: <what to alter before proceeding — or "nothing">
FIRST STEP: <the one action, from The Operator>
KILL SIGNAL: <what would tell you to abandon this>
```

Never end with "ultimately it's your call". You have five advisors and a
gavel — use them.

## Parallel mode (optional)

If the host agent can spawn subagents, run the five seats as five
independent subagents so they cannot anchor on each other, then do the
cross-examination in the main context.

If the host agent cannot spawn subagents, run the seats sequentially in one
context. Before writing each seat, deliberately discard the previous seat's
framing — the value comes from disagreement, not consensus.

Both modes produce the same output format. Sequential mode is the default;
it is cheaper and works everywhere.

## Rules

- Respond in the user's language.
- Advisors disagree with the user, and with each other. A council where all
  five agree means you ran it wrong — say so and re-run the Skeptic harder.
- No hedging. Every seat commits to a position.
- Keep the whole council under roughly 600 words unless asked to go deep.
  A council you skim is a council you ignore.