adversarial-design-review

Reviews design documents adversarially to produce severity-graded findings reports.

1|Updated Dec 13, 2025
One-click install
npx skills add https://github.com/ZK-Theory/TDL --skill adversarial-design-review-zk-theory
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: adversarial-design-review
Source: https://github.com/ZK-Theory/TDL/tree/main/.agents/skills/adversarial-design-review
Command: npx skills add https://github.com/ZK-Theory/TDL --skill adversarial-design-review-zk-theory

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Design documents, specifications, and plans often contain unsupported claims, circular evidence, missing controls, and cross-spec inconsistencies that pass superficial review. This Skill conducts an independent adversarial review that attacks the document for material errors and produces an auditable, severity-graded findings report. ## Core Features & Use Cases - Five Review Lanes: Evidence fidelity, per-component attack, cross-spec consistency matrix, currency checks against live state, and practicality/proportionality assessment. - Structured Findings Standard: Every finding carries ID, severity (Critical/Major/Minor), precise claim, file-and-line evidence, failure scenario, impact, disposition, and affected decisions. - Implementation-Conformance Mode: Reviews committed code against its accepted plan, binding every claimed control to the line that implements it. - Use Case: Before a design sign-off gate, run this Skill on a specification to catch stale citations, unenforced invariants, and untestable interfaces, then deliver a verdict of accept, accept_with_required_changes, or rework_required. ## Quick Start Use the adversarial-design-review skill to attack the attached design specification and produce a severity-graded findings report with a decision disposition.

Frequently Asked Questions about adversarial-design-review

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

FAQPage Schema
How do I adversarially review a design document or specification?

Work from fresh context against direct evidence rather than the document's own summaries. Apply the five review lanes: evidence fidelity, per-component attack, cross-spec consistency matrix, currency checks, and practicality assessment, then grade each finding by severity.

How to review code against an accepted plan or spec?

Use implementation-conformance mode: enumerate every control and step the governing prose claims and bind each to the line that implements it. An unimplemented or subset-implemented claim is a finding even when all tests pass, since tests exercise present behavior, not absent controls.

What severity levels does a design review findings report use?

Findings are graded Critical, Major, or Minor. Critical covers corrupted authority, invalid acceptance, or impossible recovery; Major covers material ambiguity, missing controls, and untestable interfaces; Minor covers local inconsistency and useful hardening.

When should I use adversarial design review versus result triage?

Use this Skill for design documents, specifications, plans, and proposals, including code reviewed against an accepted plan. For validity of a computational result or claim, use research-assurance-triage instead; for prose quality, use humanizer.

What are the limitations of an adversarial design review?

The review cannot reverse owner-approved decisions, alter human authority, migrate evidence, or authorize implementation; those escalate to the owner. When source evidence is missing or contradictory, it reports Partial rather than inferring gaps.