---
title: "Rude Q&A agent skill for pressure-testing decisions | Swipe"
description: "Rude Q&A is an adversarial agent workflow that challenges plans, replaces vague claims, and records the choices and consequences that survive."
canonical: "https://swipe.md/tools/rude-qa"
language: "en"
---

Decision making · Checked 12 August 2026

# Rude Q&A: Turn a fuzzy plan into a decision with a cost

Attack a real proposal until its choices become defensible.

Rude Q&A is an adversarial agent workflow that challenges plans, replaces vague claims, and records the choices and consequences that survive.

Type

skill

Works on

Claude Code, Agent skills, Markdown

[Open Rude Q&A ↗︎](https://github.com/asmartbear/asb-skills/blob/main/.claude/skills/asb-rude-qa/SKILL.md)

## Pressure-test something you may actually change

Rude Q&A turns an agent into a constructive hostile interviewer for one plan, decision, pitch, market, or pricing choice. It attacks vague language, checks whether the opposite choice could also be rational, and makes the user name the downside they are accepting.

The skill’s strongest rule is that discomfort is not a reason to move on. If an answer is vague or wishful, the agent stays on it and asks again. A choice only survives when both sides are credible, the negative consequence is explicit, and it does not contradict the other decisions in the plan.

The work is captured in a Markdown decision record across four phases: frame the target, interrogate it, reach a conclusion, and record what changed. The allowed outcomes are a sharpened plan, a plan that needs more evidence, or a bad idea. “Approximately fine” is deliberately excluded.

## Use one real decision

Start with a decision that is still reversible and that you are willing to change. State the proposal, what is already fixed, the worst cost of being wrong, and who else is affected. Then let the agent challenge one point at a time instead of answering a large questionnaire in one pass.

We used the method on Mailroom’s real outbound boundary. It forced a choice between faster autonomous sending and slower human approval. Mailroom kept human approval because one bad message can damage sender trust; agents may read, search, classify, and draft, but they do not silently send.

The workflow builds on Scott Berkun’s Rude Q&A exercise and packages the method as a CC BY 4.0 skill. It writes a working document, so choose its location carefully when the plan contains confidential commercial or personnel details.

## Defensible is not necessarily correct

The skill can expose contradictions and make trade-offs explicit. It cannot manufacture customer evidence, market data, or technical facts that are not in the room. A crisp plan can still be wrong.

Use the resulting open questions as research jobs. When the argument depends on what customers do, what a competitor shipped, or what the system supports, verify that evidence outside the interrogation before committing to the plan.

## Featured in Swipe

-   [Search the work your AI agents already did](https://swipe.md/issues/search-your-agent-history)

## Sources

-   [Rude Q&A skill](https://github.com/asmartbear/asb-skills/blob/main/.claude/skills/asb-rude-qa/SKILL.md)
-   [Rude Q&A source repository](https://github.com/asmartbear/asb-skills)
-   [Scott Berkun's original Rude Q&A exercise](https://scottberkun.com/2014/how-to-run-a-rude-qa/)