addyosmani

addyosmani / adverse

Public

Multi-agent adversarial code review for any coding agent

15
1
100% credibility
Found May 15, 2026 at 15 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

Adverse is a tool that simulates multiple expert reviewers using AI personas to perform thorough, cross-validated code reviews, catching logic errors, security risks, and design issues.

How It Works

1
🔍 Discover Adverse

While using AI helpers to write code, you learn about a smart tool that reviews it from three expert viewpoints to spot bugs, risks, and improvements.

2
đŸ“„ Set It Up

Quickly add the review tool to your computer or connect it to your AI coding companion, whichever feels right for you.

3
Pick Your Way
đŸ’»
Standalone Checker

Grab the tool and launch reviews directly on any code folder anytime.

🔗
AI Add-on

Plug it into your AI coding helper for reviews right within your workflow.

4
📁 Choose Code to Review

Tell the tool which folder of code or recent changes you want it to examine.

5
⏳ Reviewers Go to Work

Three specialized reviewers—auditor, adversary, and pragmatist—examine your code separately, then cross-check each other's discoveries for solid agreement.

6
📊 Get Your Report

Receive a beautiful summary with a ship-or-fix verdict, highlighted issues by severity, and clear explanations of what to watch.

✅ Code Gets Better

Armed with trustworthy insights, you fix problems confidently and ship higher-quality code.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 15 to 15 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is adverse?

Adverse runs multi-agent adversarial code reviews on your codebase using any AI coding agent, pitting three personas—Auditor for logic bugs, Adversary for security risks, and Pragmatist for maintainability—against each other in iterative debates. It ships as a Node.js CLI that wraps stdin/stdout tools like Claude Code, Codex CLI, Gemini, or Ollama, or as a native Claude Code Skill for seamless integration. You feed it a directory, git diff, or uncommitted changes, and get a synthesized report with verdicts like "SHIP-WITH-CAVEATS" plus cross-validated findings.

Why is it gaining traction?

Unlike single-shot AI reviews or pricey multi-model setups, adverse leverages one model across orthogonal personas with a cross-examination round for consensus, cutting costs and bias while flagging issues others miss—like proactive false data injection in adversarial multi-agent systems. The CLI's flexibility (`adverse review ./src --agent "ollama run llama3.1"`) and outputs (Markdown, JSON, HTML) make it dead simple for CI gates or local checks, with exit codes for scripting.

Who should use this?

AI-assisted devs on Claude Code, Aider, or Ollama who ship code daily and need rigorous pre-merge reviews. Teams building multi-agent frameworks or LLMs wanting adversarial multi-agent evaluation to catch edge cases in trust boundaries. Solo contributors reviewing PR diffs before pushing.

Verdict

Worth a test drive for adversarial multi-agent code review workflows—excellent docs, full test coverage, and zero deps make it production-ready despite 15 stars and 1.0% credibility score. Still early; pair with human eyes until it matures.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.