Multi-Agent Red Teaming System

A production-ready red teaming platform for Pi with 10 specialized security agents that coordinate to identify, exploit, and report vulnerabilities.

📦 Installation

From npm (Recommended)

# Install globally
pi install npm:pi-redteam

# Install with specific version
pi install npm:pi-redteam@1.0.2

# Install project-local
pi install -l npm:pi-redteam

# Try without installing
pi -e npm:pi-redteam

From Source

git clone https://codeberg.org/canoozie/redteam.git
cd redteam
npm install && just build
pi install -l file:.

Verify Installation

# Check extension loaded
pi  # Look for: "Red team extension loaded"

# Test commands
/attack status
/skill:redteam

🎯 Overview

This system implements multi-agent red teaming where specialized agents work together:

Advantage	Single Agent	Multi-Agent Team
Specialization	Shallow	Deep in each domain
Parallelism	Sequential	Concurrent
Exploit Chaining	Limited	Dedicated agent
Cognitive Load	Overwhelmed	Distributed

🏗️ Architecture

threat-modeler (Orchestrator)
    ↓
    ├─ Attack Agents (parallel)
    │   ├─ input-fuzzer
    │   ├─ injection-specialist
    │   ├─ auth-authorization
    │   ├─ business-logic
    │   ├─ dependency-supply-chain
    │   ├─ network-api
    │   └─ config-auditor
    │
    └─ Coordination Agents
        ├─ exploit-chainer
        └─ report-triage

📋 10 Specialized Agents

Core Attack Agents (5)

Agent	Focus
input-fuzzer	Malformed inputs, edge cases, encoding tricks
injection-specialist	SQLi, NoSQL, command, LDAP, SSTI, prompt injection
auth-authorization	Auth bypass, IDOR, privilege escalation, JWT
business-logic	Workflow abuse, race conditions, state attacks
dependency-supply-chain	CVE scanning, typosquatting, supply chain

Infrastructure Agents (2)

Agent	Focus
network-api	Exposed endpoints, CORS, rate limiting, info leakage
config-auditor	Secrets, cloud misconfigs, IaC security

Coordination Agents (3)

Agent	Focus
threat-modeler	STRIDE modeling, work packages, orchestration
exploit-chainer	Chains vulns into attack paths
report-triage	Deduplication, CVSS scoring, reporting

🚀 Quick Start

Start a Red Team Operation

# Initialize with target
/attack start https://target.com

# Check status
/attack status

# Progress through phases
/attack phase recon
/attack phase exploit
/attack phase reporting

# Generate report
/report

Use Security Scripts

# Load skill
/skill:redteam

# Run checks
./scripts/header-analyzer.sh https://target.com
./scripts/dependency-check.sh /path/to/project
./scripts/port-scan.sh 192.168.1.1
./scripts/sstl-check.sh example.com

Team Configurations

# Full team (all 10 agents)
pi --team full-team

# Attack team (7 agents)
pi --team attack-team

# Recon team (3 agents)
pi --team recon-team

# Quick test (3 agents)
pi --team quick-test

📊 Example Output

Vulnerability Finding

{
  "id": "AUTH-001",
  "title": "IDOR in User Profile",
  "severity": "high",
  "cvss_score": 7.4,
  "cwe": "CWE-639",
  "description": "Application fails to verify user owns requested resource",
  "proof_of_concept": "GET /api/users/123/profile with user_456's session",
  "recommendation": [
    "Add object-level authorization check",
    "Use indirect object references"
  ]
}

Exploit Chain

{
  "chain_id": "CHAIN-001",
  "name": "Account Takeover",
  "severity": "critical",
  "cvss_score": 9.0,
  "steps": [
    "1. Enumerate usernames (Info Disclosure)",
    "2. Brute force passwords (No Rate Limiting)",
    "3. Access victim data (IDOR)"
  ],
  "impact": "Complete account compromise"
}

🛡️ Safety & Ethics

Requirements

• ✅ Written authorization from system owner
• ✅ Clearly defined scope
• ✅ Approved testing windows
• ✅ Emergency contacts identified

Built-in Safeguards

1. Scope Enforcement: Agents flag out-of-scope actions
2. Destructive Blocking: Extension blocks dangerous commands
3. Audit Logging: All actions logged
4. Environment Detection: Identifies staging vs production

Code of Conduct

DO:

• ✅ Test only authorized systems
• ✅ Stay within defined scope
• ✅ Use staging when possible
• ✅ Report findings responsibly

DO NOT:

• ❌ Test without authorization
• ❌ Exceed scope
• ❌ Modify production data
• ❌ Conduct DoS without approval

📁 File Structure

.pi/
├── agents/
│   ├── teams.yaml              # 8 team configurations
│   ├── threat-modeler.md       # Orchestrator
│   ├── input-fuzzer.md         # Input testing
│   ├── injection-specialist.md # Injection attacks
│   ├── auth-authorization.md   # Auth testing
│   ├── business-logic.md       # Logic abuse
│   ├── dependency-supply-chain.md
│   ├── network-api.md          # Network security
│   ├── config-auditor.md       # Config auditing
│   ├── exploit-chainer.md      # Exploit chains
│   └── report-triage.md        # Reporting
└── skills/
    └── redteam/
        ├── SKILL.md
        ├── scripts/            # 4 security scripts
        ├── references/         # OWASP, STRIDE, checklist
        └── assets/             # Security baseline

extensions/
└── redteam.ts                  # Custom extension

🔧 Customization

Add Custom Agents

---
name: custom-agent
description: Your custom agent
tools: read,bash,grep,find,ls
---
Your system prompt...

Then add to .pi/agents/teams.yaml:

custom-team:
  - threat-modeler
  - custom-agent
  - report-triage

Extend the Extension

Add tools to extensions/redteam.ts:

pi.registerTool({
  name: "custom_tool",
  label: "Custom Tool",
  description: "Your tool",
  parameters: Type.Object({ /* schema */ }),
  async execute(toolCallId, params, signal, onUpdate, ctx) {
    return { content: [{ type: "text", text: "Result" }], details: {} };
  }
});

📚 Documentation

• README.md - This file (installation & usage)
• Agent files - Detailed agent documentation
• SKILL.md - Skill usage and scripts
• CONTRIBUTING.md - Development guide
• CHANGELOG.md - Version history
• PUBLISHING.md - npm publishing guide

🤝 Contributing

# Setup development
git clone https://codeberg.org/canoozie/redteam.git
cd redteam
npm install

# Build and test
just build
just test-local

# See all commands
just help

See CONTRIBUTING.md for details.

📦 Package Info

• Name: pi-redteam
• Version: 1.0.2
• License: MIT
• Size: ~50 KB (gzipped)
• Keywords: pi-package, security, redteam, pentest

Dependencies

Peer (provided by Pi):

• @mariozechner/pi-coding-agent
• @mariozechner/pi-ai
• @mariozechner/pi-tui
• @sinclair/typebox

Runtime: None (self-contained)

📚 References

• OWASP Top 10
• OWASP Testing Guide
• STRIDE
• CVSS Calculator
• PortSwigger Academy

⚠️ Disclaimer

This tool is for authorized security testing only. The authors accept no liability for misuse. Unauthorized testing is illegal.

---

Built for Pi | Multi-Agent Red Teaming | Production-Ready

For issues: Codeberg Issues