Adversarial Review
Have Codex try to break a change before it ships — a skeptical review that challenges the design, assumptions, and failure modes of a diff or branch and returns structured findings (verdict, severity, file:line, confidence). Trigger on "adversarial review", "challenge this change/design with Codex", "pressure-test this before I ship", "have Codex try to break this", "is this safe to merge", or a review focused on a risk area (auth, data loss, rollback, race conditions, retries). Do NOT use for a routine bug/style pass over a diff (that's diff-review), for critiquing a plan before code exists (
Install
/plugin marketplace add robbyt/claude-skillsThe repo is a Claude Code plugin marketplace. Run /plugin afterwards to pick the plugin that includes this skill.
npx skills add robbyt/claude-skills --skill adversarial-reviewgit clone --depth 1 https://github.com/robbyt/claude-skills.git /tmp/claude-skills
mkdir -p ~/.claude/skills
cp -r /tmp/claude-skills/plugins/codex/skills/adversarial-review ~/.claude/skills/adversarial-reviewPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Adversarial Review via Codex
Have Codex argue against shipping a change: question the chosen approach, the assumptions it relies on, and how it fails under real conditions. This is different from `diff-review`, which looks for defects in the code as written. Codex consults; Claude writes.
Prompt and output schema adapted from OpenAI's Codex plugin for Claude Code (openai/codex-plugin-cc, Apache-2.0).
## Transport
**Always use the MCP tool.** The plugin's MCP server (`scripts/codex-mcp-server.mjs`) runs `codex exec` for each call. Tool name: `mcp__plugin_codex_cli__codex`. If it errors with an unknown-tool error, run `/mcp` and substitute the actual prefix.
## Model
**Pin `model: "gpt-6.2-sol"` with `config: { "model_reasoning_effort": "medium" }`.** Don't downgrade to `gpt-6-luna`; this review depends on reasoning depth. Honor an explicit user-named model or effort. See `../references/patterns.md` → Models.
## Flow
1. **Pick the target and save the diff** in the project root (Codex reads files relative to it):
```bash
git diff HEAD > codex-review.diff # uncommitted work (staged + unstaged)
git diff main...HEAD > codex-review.diff # branch vs base
```
`git diff HEAD` doesn't include untracked files. If `git status --short` shows new files that belong to the change, `git add -N <file>` them first so they appear in the diff. If the diff is empty, tell the user there is nothing to review and stop.Details
- Skill name
adversarial-review- Repository
- robbyt/claude-skills
- Path
plugins/codex/skills/adversarial-review/SKILL.md- Repo updated
- 18 November 2026
FAQ
What does the Adversarial Review skill do?
Have Codex try to break a change before it ships — a skeptical review that challenges the design, assumptions, and failure modes of a diff or branch and returns structured findings (verdict, severity, file:line, confidence). Trigger on "adversarial review", "challenge this change/design with Codex", "pressure-test this before I ship", "have Codex try to break this", "is this safe to merge", or a review focused on a risk area (auth, data loss, rollback, race conditions, retries). Do NOT use for a routine bug/style pass over a diff (that's diff-review), for critiquing a plan before code exists (
How do I install the Adversarial Review skill?
Claude Code plugin: /plugin marketplace add robbyt/claude-skills. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Adversarial Review popular?
It lives in robbyt/claude-skills, which has 42 GitHub stars and 2 forks.
Related skills
Actions
42robbyt/claude-skills
GitHub Actions workflow management using GitHub CLI. Trigger when user wants to check CI status ("check CI", "is the build passing"), view workflow runs ("show recent runs", "view action logs"), watch running workflows ("watch CI"), or rerun failed workflows ("rerun the build").
API Review
42robbyt/claude-skills
This skill should be used when reviewing API designs, auditing REST or gRPC endpoints, designing backend APIs, evaluating BFF architecture, or discussing API best practices. Also triggers on: "review this API", "API design review", "is this API correct", "REST API audit", "gRPC API review", "PATCH vs PUT", "etag", "idempotency", "pagination design", "API error handling", "backward compatibility", "BFF architecture", "resource-oriented design", "TOCTOU", "race condition", "optimistic concurrency", "field mask", "long-running operation", "custom method", "soft delete", "API naming conventions",
Artifact Detect
42robbyt/claude-skills
This skill should be used when the user asks "what's wrong with this video", "why does this video look bad", "detect video artifacts", "find quality issues", "video has artifacts", "identify compression artifacts", or wants to diagnose specific quality problems in a video file.
Test Driven Development
43lifangda/claude-plugins
Use when implementing any feature or bugfix, before writing implementation code - write the test first, watch it fail, write minimal code to pass; ensures tests actually verify behavior by requiring failure first
Codebase Analysis
42robbyt/claude-skills
Deep architectural analysis of the current workspace using Google Antigravity (`agy`). Trigger when the user needs an architecture overview ("analyze this codebase with agy", "map dependencies with Google Antigravity"), is onboarding to unfamiliar code, exploring legacy systems, or hunting technical debt. Replaces the deprecated gemini-cli `codebase_investigator` workflow.
Hard Call
40ArogyaReddy/alirezarezvani-claude-skills
/em -hard-call — Framework for Decisions With No Good Options
Browse the most popular Claude skills, all software development skills, or write your own with the skills generator.