skills.rest
nkootstra
nkootstraCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

adversarial-review

Coordinate multi-agent adversarial review to identify and verify code issues.

Updated Aug 27, 2026
Search Tags:#verification#reporting#pipeline#review#code#adversarial#bug
One-click install
npx skills add https://github.com/nkootstra/skills --skill adversarial-review-nkootstra
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: adversarial-review
Source: https://github.com/nkootstra/skills/tree/main/skills/adversarial-review
Command: npx skills add https://github.com/nkootstra/skills --skill adversarial-review-nkootstra

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill enables teams to perform robust code reviews by orchestrating a three-agent adversarial pipeline that surfaces true issues while minimizing false positives.

Core Features & Use Cases

  • Orchestrates the Bug Finder, Adversary, and Referee roles to generate diverse findings and converge on truth.
  • Supports both CLI (Claude Code / Codex) parallel subagents and Claude.ai sequential simulation, with clearly separated prompts and clean contexts.
  • Produces a structured final report with per-item IDs, severities, verdicts, reasoning, and a consolidated referee verdict.

Quick Start

Run the three-agent adversarial review pipeline on your codebase to generate a final verdict report.

Frequently Asked Questions about adversarial-review

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

FAQPage Schema
How does an adversarial code review pipeline work?▼

Yes, multi-agent adversarial review can run in Claude.ai using sequential simulation. This mode cleanly separates prompts and contexts to emulate the Bug Finder, Adversary, and Referee roles without requiring parallel execution environments.

Can I automate code review to reduce false positives in bug reporting?▼

To run adversarial code review, apply the pipeline to your codebase in CLI mode. The tool orchestrates real subagents in parallel to identify bugs, challenge findings, and generate a final verdict report with bug IDs, locations, and severities.

Does the adversarial review pipeline support CLI mode with parallel subagents?▼

The pipeline produces a final structured report detailing bug IDs, severities, code locations, rationales, and referee verdicts. This structured format ensures teams receive a consolidated, high-confidence assessment of verified code issues.

What is the best way to verify code issues with high confidence during a review?▼

The adversarial review pipeline is designed for robust codebase analysis. It is best suited for code-review tasks where identifying true bugs and minimizing false positives through multi-agent verification is critical for pipeline reporting.

When should I use a multi-agent pipeline instead of a standard code review?▼

Yes, the adversarial review pipeline operates with no dependencies. It supports both CLI parallel subagents and Claude.ai sequential simulation, making it accessible for diverse code-review tasks without requiring external modules.

What format does the final code review report use for bug tracking?▼

The adversarial pipeline coordinates three roles: a Bug Finder to surface issues, an Adversary to challenge them, and a Referee to converge on truth. This multi-agent interaction ensures robust verification and structured bug reporting.

Related Skills

Explore More Software Engineering Skills→
SensLiaoSensLiao

workflow-creator

Create JavaScript orchestrator files for deterministic multi-agent workflows in Claude Code.

Community
Advanced
githubnextgithubnext
4.9k

temporary-id-safe-output

Resolve temporary IDs to GitHub issue numbers within agent-driven workflows.

Official
Basic
phuhao00phuhao00
12

code-analyst-agent

Analyze source code to identify symbols, call graphs, and architectural patterns.

Community
Intermediate
Eunho-JEunho-J
2

ecomode

Configure an AI operational mode to reduce computational overhead and API costs.

Community
Basic