Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
Parth Maheshwari avatar

Adversarial Dev

  • 1 repo stars
  • Updated April 12, 2026
  • parthpm/adversarial-dev-plugin

Plan and build features with Codex adversarial review at every step.

About

adversarial-dev is a Claude Code skill in the Productivity & Planning category. Plan and build features with Codex adversarial review at every step.

  • adversarial-dev
  • Productivity & Planning
  • AI-coding skill

Adversarial Dev by the numbers

  • Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add parthpm/adversarial-dev-plugin
/plugin install adversarial-dev@parthpm-adversarial-dev

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
repo stars1
Last updatedApril 12, 2026
Repositoryparthpm/adversarial-dev-plugin

What it does

Plan and build features with Codex adversarial review at every step.

README.md

adversarial-dev

Plan and build with cross-model adversarial review.

A Claude Code plugin that adds two workflow skills — /adversarial-dev:plan and /adversarial-dev:build — backed by Codex adversarial review. A different model (GPT via Codex) reviews Claude's output at every step, catching what same-model review cannot.

How It Works

Most AI code review is same-model: Claude reviews Claude's output. This has a blind spot — the reviewer shares the same biases, training artifacts, and failure modes as the author.

Cross-model adversarial review breaks this by routing review through an independent model (GPT via Codex CLI). The reviewer has different strengths, different blind spots, and no shared context with the author — so it catches assumptions, logic errors, and missed edge cases that same-model review systematically misses.

┌─────────────────────────────────────────────────────────────┐
│                    /adversarial-dev:plan                     │
│                                                             │
│  Explore ──► Draft Plan ──► Codex Review ──► Simplify       │
│                              ▲       │                      │
│                              └───────┘                      │
│                            iterate until clean              │
│                                    │                        │
│                              plan file (plans/)             │
└────────────────────────────────────┬────────────────────────┘
                                     │
                                     ▼
┌─────────────────────────────────────────────────────────────┐
│                   /adversarial-dev:build                     │
│                                                             │
│  Read Plan ──► Implement ──► Test ──► Codex Review          │
│                                        ▲       │            │
│                                        └───────┘            │
│                                      iterate until clean    │
└─────────────────────────────────────────────────────────────┘

The plan phase produces a file. The build phase consumes it. Both phases are independently pressure-tested by Codex adversarial review before proceeding.

Prerequisites

Installation

Note: The official Codex plugin blocks programmatic invocation of /codex:adversarial-review (openai/codex-plugin-cc#156 fixes this). Until that PR merges, install from the fork below.

Step 1: Install Codex plugin (fork with fix)

/plugin marketplace add parthpm/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
/codex:setup

Step 2: Install adversarial-dev plugin

/plugin marketplace add parthpm/adversarial-dev-plugin
/plugin install adversarial-dev@parthpm-adversarial-dev
/reload-plugins

After openai/codex-plugin-cc#156 merges

Switch to the official Codex source:

/plugin marketplace add openai/codex-plugin-cc
/plugin update codex@openai-codex

The plugin identity (openai-codex) is the same in both the fork and official repo, so this is a seamless switch.

Usage

Plan a change

/adversarial-dev:plan add rate limiting to the /api/upload endpoint

This explores the codebase, drafts an implementation plan, pressure-tests it with Codex adversarial review, simplifies, and presents the result. No code is modified — only a plan file is written to plans/.

Build from a plan

/adversarial-dev:build

With no arguments, it picks up the most recent plan from plans/ or .claude/plans/. You can also give it a task directly:

/adversarial-dev:build fix the race condition in the connection pool

This implements the changes, runs the full test suite, pressure-tests with Codex adversarial review, and iterates until clean. Changes are left uncommitted for your review.

Known Limitation

The Codex plugin's adversarial-review skill has disable-model-invocation: true, which prevents programmatic invocation. PR openai/codex-plugin-cc#156 removes this restriction. Until it merges, use the fork marketplace (parthpm/codex-plugin-cc) as shown in the installation instructions above.

License

MIT

Related skills

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.