# Brutal Review

> Adversarially review an idea, plan, design, implementation, or draft before commitment. Use when the user asks for a brutal review, stress test, pre-mortem, hostile reviewer, failure analysis, or a candid go/no-go judgment. Do not use for ordinary proofreading or when the user only wants implementation.

- Skill: `denizaslan36/brutal-review` (Agent Skill)
- Install (CLI): `npx skillmds@latest add denizaslan36/brutal-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/denizaslan36/brutal-review/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: denizaslan36 (https://skillmd.com/u/denizaslan36)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/denizaslan36/brutal-review

---


# Brutal Review

Find the cheapest way to expose a costly mistake before it ships.

## Method

1. Identify the artifact's intended outcome, audience, and non-negotiable constraints from the available context. Ask only for missing information that could reverse the judgment.
2. Attack the strongest plausible version of the artifact, not a weakened caricature.
3. Inspect, in order:
   - false or hidden assumptions;
   - contradictions and missing dependencies;
   - failure modes and edge cases;
   - cost, complexity, and maintenance burden;
   - claims that lack direct evidence;
   - simpler alternatives that achieve the same outcome.
4. Rank findings by expected damage and likelihood. Do not inflate a stylistic preference into a blocker.
5. Make each material finding traceable to the precise weakness, supporting
   evidence, likely impact, and cheapest adequate correction. Use explicit
   **Finding / Evidence / Impact / Cheapest correction** labels when there are
   multiple findings or the decision is high-risk; for a simple case, combine
   them into concise prose instead of repeating a template.
6. End with one verdict: **go**, **go with conditions**, **revise**, or **stop**. State the decisive reason in one sentence.

## Discipline

- Separate observed facts, inferences, and unknowns.
- Do not manufacture objections to appear rigorous.
- Do not soften a material defect with generic praise.
- Do not propose a rewrite unless the user asks for one.
- If evidence is insufficient, say what remains unknown rather than guessing.