adversarial-review

Generate prioritized, evidence-backed findings from dual Claude and Codex document reviews.

Updated Mar 22, 2026
One-click install
npx skills add https://github.com/engineai-nz/engineai-skills --skill adversarial-review-engineai-nz
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: adversarial-review
Source: https://github.com/engineai-nz/engineai-skills/tree/main/review
Command: npx skills add https://github.com/engineai-nz/engineai-skills --skill adversarial-review-engineai-nz

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It exposes gaps, contradictions, missing sections, unrealistic claims, and fragile architecture/metrics in PRDs, architecture docs, SOWs, and technical specs before they become expensive to fix.

Core Features & Use Cases

  • Dual-agent adversarial review: Runs independent reviewers (Claude + Codex) and synthesises results into a single report.
  • Evidence-driven findings: Produces prioritized P0–P3 findings with required fields, rationale, and suggested fixes.
  • Structured outputs for actionability: Generates a full internal report, a send-ready author brief, and a forward-compatible JSON artifact.
  • Proof Burden Mode for AI-ish documents: Automatically forces explicit answers on evaluation methods, fallbacks, human approvals, trust boundaries, and cost/latency when AI-generation signals appear.
  • Independent Codex fact-checking pass: Verifies the opponent’s claims independently and surfaces disagreements.

Quick Start

Use adversarial-review to review the currently-open document by asking for a dual-agent adversarial critique that outputs a prioritized, evidence-backed report and an author brief.

Frequently Asked Questions about adversarial-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I stress-test a technical spec for gaps and contradictions?

Stress-testing a technical spec requires an adversarial review that uses dual-agent scrutiny to expose gaps, contradictions, and fragile architecture, generating prioritized evidence-backed findings before they become expensive to fix.

What is dual-agent adversarial document review?

Dual-agent adversarial review is a mechanism where independent Claude and Codex passes critique a document, verify claims with citations, and synthesize prioritized findings into a single report and author brief.

Can I use adversarial review to fact-check a PRD?

Yes, you can perform a PRD critique by extracting goals, assumptions, and metrics, then running an independent fact-checking pass to surface disagreements, unrealistic claims, and missing sections.

Does this approach work for evaluating AI feature specs?

Yes, evaluating AI feature specs triggers a proof burden mode that forces explicit answers on evaluation methods, fallbacks, human approvals, trust boundaries, and cost or latency constraints.

What document types are supported for architecture fitness evaluation?

Architecture fitness evaluation and metrics validation apply to PRDs, architecture docs, statements of work, technical specs, and GTM or strategy docs when checking for fragile claims.

How do I get structured outputs from a document critique?

A document critique generates structured outputs by synthesizing findings into a full markdown report, a send-ready author brief, and a forward-compatible JSON artifact containing prioritized P0 to P3 issues.