Adversarial Code Reviewer — Development

Adversarial Code Reviewer forces a genuinely critical review of code changes by adopting three hostile reviewer personas (Saboteur, New Hire, Security Auditor). Each persona must find at least one issue, eliminating rubber-stamp approvals. It produces a structured report with severity-classified findings and a BLOCK/CONCERNS/CLEAN verdict.

engineering claude-code

📂 View skill on GitHub ↗ 🚀 How to use

✨ What it does

Three adversarial personas (Saboteur, New Hire, Security Auditor) each with distinct priorities
Mandatory findings: each persona must surface at least one issue, preventing rubber-stamp reviews
Severity promotion: issues caught by multiple personas are promoted one severity level
Self-review trap breaker: techniques like bottom-up reading and contract checking
Structured verdicts: BLOCK, CONCERNS, or CLEAN with clear merge guidance

🎯 When to use it

Before merging any PR, especially self-authored PRs with no human reviewer
After a long coding session when fatigue may cause blind spots
When Claude or another reviewer gave an easy 'looks good' approval
On security-sensitive code such as auth, payments, data access, or API endpoints
When something 'feels off' about the code and you want a second opinion

🚀 How to use

Trigger: /adversarial-review

Trigger with the slash command /adversarial-review optionally followed by flags: --diff <ref> to review a range of commits, --file <path> to review a specific file, or no arguments to review staged/unstaged changes. The skill expects a git repository and will gather the diff or file content automatically. Example prompts:

/adversarial-review
/adversarial-review --diff HEAD~3
/adversarial-review --file src/auth.ts

📄 Output: A structured markdown report with scope, verdict, critical findings, warnings, notes, and a summary.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering-team/skills/adversarial-reviewer .claude/skills/adversarial-reviewer

Skill source: engineering-team/skills/adversarial-reviewer/SKILL.md

⚠️ Good to know

Requires a git repository to review changes; if no changes are found, it stops and reports 'Nothing to review.'

❓ FAQ

What happens if a persona finds no issues?

Each persona MUST find at least one issue; if none is found, the persona must go back and look harder, noting the most fragile assumption or likely point of confusion.

How are findings from multiple personas handled?

Duplicate findings are merged, and any issue caught by 2+ personas is promoted one severity level (e.g., NOTE becomes WARNING).

What does the verdict CLEAN mean?

CLEAN means only NOTE-level findings were found, so the code is considered safe to merge.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.