adversarial-review-cycle

Run two-pass adversarial reviews with severity classification and ship-or-spin gating.

Updated Aug 23, 2026

Search Tags:#security #test coverage #triage #hardening #adversarial review #code diff #release readiness

Ready to run

npx skills add https://github.com/ElvinOuyang/claude-skill-collection --skill adversarial-review-cycle

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: adversarial-review-cycle
Source: https://github.com/ElvinOuyang/claude-skill-collection/tree/main/plugins/workflow-toolkit/skills/adversarial-review-cycle
Command: npx skills add https://github.com/ElvinOuyang/claude-skill-collection --skill adversarial-review-cycle

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It prevents avoidable production failures by forcing an explicitly hostile, adversarial review loop that hunts for bugs, security gaps, and edge cases the implementer is likely to miss.

Core Features & Use Cases

Two-pass adversarial hardening: Run a Round 1 broad attack and then a Round 2 review from a different angle (e.g., security, correctness, UX, or performance) before considering the work ready.
Full-diff, test-inclusive scrutiny: Require the reviewer to read the full git diff (or read each changed file end-to-end) and to read the tests themselves, not rely on summaries.
Severity-based triage and gating: Classify findings as P0/P1/P2/P3, fix P0/P1 before proceeding, and only declare “ship clean” when both rounds return acceptably clean results (with explicit rationale for accepted lower-severity issues).

Quick Start

Run an adversarial review cycle on your completed changes by asking for a “hardening pass” in explicitly hostile terms like “Poke holes in this and try to break it before I ship.”

Frequently Asked Questions about adversarial-review-cycle

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

FAQPage Schema

What is an adversarial code review and how does it prevent production failures?▼

An adversarial code review is a structured hardening pass that hunts for bugs and security gaps the implementer missed. It forces an explicitly hostile review loop to identify and fix P0/P1 issues before release, preventing avoidable production failures.

How do I run an adversarial hardening pass on my completed code changes?▼

To run an adversarial hardening pass, prompt the reviewer in explicitly hostile terms like “Poke holes in this and try to break it.” The cycle requires two review rounds from different angles, full-diff consumption, and severity classification before a ship-or-spin decision.

Can I use this adversarial review process for iOS, web, and AI agent workflows?▼

Yes, you can use this adversarial review process for completed, non-trivial iOS, web, and AI agent changes. The work must be stable enough for a full-diff and test read-through to identify and fix P0/P1 issues before release.

What is the best way to triage findings during a security hardening review?▼

The best way to triage findings during a security hardening review is to classify them by severity as P0, P1, P2, or P3. You must fix all P0 and P1 issues before proceeding, and only declare “ship clean” when both review rounds return acceptably clean results.

Does a full-diff adversarial review require reading the tests themselves?▼

Yes, a full-diff adversarial review requires the reviewer to read the tests themselves, not rely on summaries. The reviewer must read the full git diff or each changed file end-to-end alongside the tests to ensure comprehensive scrutiny.

When should I block a release after an adversarial review cycle?▼

You should block a release after an adversarial review cycle if there are unresolved P0 or P1 issues. The ship-or-spin decision blocks on these high-severity findings, requiring explicit rationale for accepting any lower-severity P2 or P3 issues before shipping.