adversarial-workflow

11 adversarial multi-agent slash commands for Claude Code — covering the full software development lifecycle. Each workflow pits multiple independent AI agents against each other to produce higher-quality, battle-tested output.

How it works

Traditional AI coding: one agent proposes → you accept.

Adversarial: attacker agents propose solutions → verifier agents independently scrutinize each one → only outputs that survive verification reach you. No rubber-stamping.

User request
    │
    ▼
┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│  Agent A    │     │  Agent B    │     │  Agent C    │  ← parallel attackers
│  (propose)  │     │  (propose)  │     │  (propose)  │
└──────┬──────┘     └──────┬──────┘     └──────┬──────┘
       └────────────────────┼────────────────────┘
                            ▼
                   ┌─────────────────┐
                   │  Verifier Pool  │  ← independent review
                   │  (reject/pass)  │
                   └────────┬────────┘
                            ▼
                      Final output

Installation

npx adversarial-workflow install

This copies all 11 workflow scripts into your project's .claude/commands/ directory, making them available as /adversarial-* slash commands inside Claude Code.

Workflows

| Command | Use case | |---|---| | /adversarial-bugfix | Fix bugs: parallel patches → adversarial verification | | /adversarial-feature | Build features: spec → design → parallel implementation → gate | | /adversarial-security | Security audit: parallel attack vectors → exploitability scoring | | /adversarial-review | PR review: 4 parallel reviewers (security/perf/correctness/style) → REVIEW.md | | /adversarial-tests | Generate tests: parallel strategies → coverage gate | | /adversarial-deps | Dependency audit: CVEs, outdated packages, license conflicts | | /adversarial-rca | Root cause analysis: parallel hypotheses → evidence ranking | | /adversarial-debt | Tech debt assessment: parallel scanners → priority-ranked backlog | | /adversarial-migration | Plan migrations: parallel path analysis → risk-weighted rollout | | /adversarial-scaffold | Greenfield build: spec → parallel scaffolding → integration gate | | /adversarial-crates | Decompose codebase into independently buildable units |

Usage

After installation, restart your Claude Code session and use any slash command:

/adversarial-bugfix   fix the authentication bypass in auth.ts
/adversarial-security audit the payment processing module
/adversarial-review   PR #142
/adversarial-tests    generate tests for UserService

Requirements

• Claude Code (any version)
• Node.js ≥ 16
• No other dependencies

Contributing

See CONTRIBUTING.md for how to add new workflows or improve existing ones.

Changelog

See CHANGELOG.md for version history.

License

MIT © 2025 Dung Nguyen