agent-wars

Multi-agent security testing CLI tool. Three AI agents battle-test your codebase:

• Red Agent - Finds security vulnerabilities
• Blue Agent - Applies automatic fixes
• Green Agent - Orchestrates and referees the process

Built with Claude Agent SDK.

Requirements

• Node.js 18+
• ANTHROPIC_API_KEY environment variable set

Install

npm install -g agent-wars

Usage

# Scan a project
agent-wars scan -t ./my-project

# Scan with options
agent-wars scan -t ./my-project -o ./report -i 5 --no-fix

# Use a specific model
agent-wars scan -t ./my-project -m claude-sonnet-4-5-20250929

Options

| Flag | Description | Default | |------|-------------|---------| | -t, --target <path> | Target project directory | (required) | | -o, --output <path> | Report output directory | ./agent-wars-report | | -i, --max-iterations <n> | Max Red/Blue rounds | 3 | | -s, --severity <level> | Min severity: critical|high|medium|low|info | low | | --fix / --no-fix | Apply automatic fixes | true | | -m, --model <model> | Claude model to use | claude-sonnet-4-5-20250929 | | -b, --budget <usd> | Max API budget in USD | 5 | | -v, --verbose | Verbose output | false |

How It Works

1. Red Agent scans your codebase for vulnerabilities (OWASP Top 10, injection, auth issues, etc.)
2. Blue Agent analyzes found vulnerabilities and applies fixes
3. Green Agent orchestrates multiple rounds until all issues are resolved or max iterations reached
4. HTML + JSON reports are generated with full details

License

MIT