Codex Red Team Mode

Skill by ara.so — Codex Skills collection.

Overview

Codex Red Team Mode is a lightweight, opt-in configuration layer that enables AI coding assistants to think like red team operators during authorized security assessments. Unlike persistent jailbreaks that pollute normal operations, this project provides:

• Opt-in activation: Normal mode by default, red team mode only when explicitly enabled
• Structured routing: phase -> router -> pack -> leaf workflow for security operations
• Session isolation: State management via JSON runtime files
• Lightweight hooks: Minimal context bloat with targeted prompt overlays

The project supports multiple security phases including web exploitation, Active Directory, post-exploitation, reverse engineering, code auditing, payload development, and evasion techniques.

Installation

The installer uses managed incremental installation that preserves your existing configuration while injecting only the necessary red team components.

Python (Cross-platform)

# Clone the repository
git clone https://github.com/chAng-L19/codex-redteam-mode.git
cd codex-redteam-mode

# Run installer
python scripts/install.py

Windows (PowerShell)

# Navigate to project directory
cd codex-redteam-mode

# Run installer with execution policy bypass
powershell -ExecutionPolicy Bypass -File .\scripts\install.ps1

macOS / Linux

# Ensure Python 3 is available
python3 scripts/install.py

The installer will:

• Preserve your original AGENTS.md and hooks.json
• Inject managed blocks from the repository
• Remove old version runtime remnants
• Write installation manifest
• Automatically run validation

Validation

After installation, verify the setup:

# Run validation script
python scripts/validate.py

# Or run full test suite
python -m unittest discover -s tests -p "test_*.py"

Core Commands

Enable Red Team Mode

# Full command
/redteam on

# Light mode (targeted security analysis)
/redteam light

# Full mode (complete red team workflow)
/redteam full

# Natural language
enable red team mode
Enter Red Team Mode

Disable Red Team Mode

# Command
/redteam off

# Natural language
disable red team mode
Exit Red Team Mode

Check Current Mode

# Query current state
what mode am I in?
show red team status

Modes

Workflow Structure

The routing mainline follows:

phase -> router -> pack -> leaf

Core Phases

• web: Web application testing (XSS, SQLi, CSRF, etc.)
• ad: Active Directory exploitation
• postex: Post-exploitation activities
• reverse: Reverse engineering
• code-audit: Source code security analysis
• payload: Payload generation and customization
• evasion: AV/EDR evasion techniques

Extended Routers/Packs

• recon: Reconnaissance and enumeration
• api: API security testing
• auth: Authentication/authorization bypass
• injection: Various injection attacks
• file: File upload/download vulnerabilities
• cloud: Cloud infrastructure security
• container: Docker/Kubernetes security
• network: Network protocol analysis
• crypto: Cryptographic vulnerabilities
• mobile: Mobile application security

Usage Patterns

Web Security Testing

# In red team mode, analyze a web application
"""
/redteam full

I need to assess a web application at https://target.example.com (authorized)
Focus on:
- Authentication mechanisms
- Session management
- Input validation
- API endpoints
"""

# The AI will route through: web -> api -> injection -> leaf
# Providing structured analysis with appropriate techniques

Active Directory Assessment

# Enable red team mode for AD testing
"""
/redteam full

Analyzing Active Directory environment (authorized pentest)
- Domain: corp.example.local
- Current access: low-privilege domain user
- Goal: privilege escalation paths
"""

# Routes through: ad -> enum -> privesc -> pack

Code Security Audit

# Light mode for code review
"""
/redteam light

Review this authentication function for security issues:

def authenticate(username, password):
    query = f"SELECT * FROM users WHERE username='{username}' AND password='{password}'"
    result = db.execute(query)
    return result.fetchone()

"""

### Payload Development

```python
# Full mode for payload creation
"""
/redteam full

Need to create a Python reverse shell payload that:
- Connects to ${C2_SERVER}:${C2_PORT} (env vars)
- Evades basic AV signature detection
- Works on Windows 10+
- Maintains persistence
"""

# Routes through: payload -> evasion -> windows -> leaf

Configuration

Runtime State Files

The mode state is stored in JSON format, typically in:

.codex/runtime/redteam_state.json

Example state file:

{
  "mode": "redteam-full",
  "phase": "web",
  "router": "injection",
  "pack": "sqli",
  "session_id": "uuid-here",
  "timestamp": "2026-05-19T12:34:56Z"
}

Hooks Configuration

The hooks.json file manages prompt overlays:

{
  "redteam": {
    "enabled": false,
    "mode": "normal",
    "phases": ["web", "ad", "postex", "reverse", "code-audit", "payload", "evasion"],
    "routing": "phase-router-pack-leaf"
  }
}

Custom Phase Extension

To add custom phases, edit the routing configuration:

# Example: Adding a custom "wireless" phase
# Edit runtime/routing.json

{
  "phases": {
    "wireless": {
      "routers": ["wifi", "bluetooth", "rfid"],
      "packs": {
        "wifi": ["wpa", "wep", "wps", "evil-twin"],
        "bluetooth": ["ble", "classic", "pairing"]
      }
    }
  }
}

Troubleshooting

Installation Issues

# Check if files were properly injected
ls -la .codex/AGENTS.md
cat .codex/hooks.json | grep redteam

# Re-run installer with verbose output
python scripts/install.py --verbose

# Clean install (removes all managed blocks first)
python scripts/install.py --clean

Mode Not Activating

# Verify runtime state
cat .codex/runtime/redteam_state.json

# Check hooks are loaded
python -c "import json; print(json.load(open('.codex/hooks.json'))['redteam'])"

# Force mode switch
/redteam off
/redteam full

Validation Failures

# Run specific test
python -m unittest tests.test_routing

# Check installer integrity
python scripts/validate.py --check-install

# Verify phase detection
python scripts/validate.py --test-phases

Context Pollution

If normal mode feels influenced by red team prompts:

# Ensure clean mode switch
/redteam off

# Verify state
python -c "import json; print(json.load(open('.codex/runtime/redteam_state.json'))['mode'])"

# Should output: "normal"

Security & Legal Considerations

⚠️ Authorization Required: Only use red team mode on systems where you have explicit written authorization.

# Always document scope
"""
AUTHORIZATION:
- Client: Example Corp
- Scope: https://test.example.com
- Date: 2026-05-20 to 2026-05-27
- Contact: security@example.com
"""

Never:

• Use on production systems without authorization
• Use on third-party systems
• Share credentials or access outside scope
• Ignore legal boundaries

Integration Examples

With MCP Tools

# Using with Model Context Protocol tools
"""
/redteam full

Using ${MCP_TOOL_NMAP} scan:
- Target: ${TARGET_NETWORK}
- Ports: 1-65535
- Service detection enabled
- Output: XML format
"""

With CI/CD Security Scanning

# Light mode for automated security review
"""
/redteam light

Analyze this PR for security issues:
- Changes in auth/* files
- New API endpoints
- Database query modifications
"""

Best Practices

1. Always start in normal mode: Only activate red team mode when needed
2. Use light mode for analysis: Reserve full mode for active testing
3. Document your scope: Keep authorization details in session context
4. Clean state between engagements: Use /redteam off when switching contexts
5. Validate after updates: Run tests after any configuration changes
6. Use environment variables: Never hardcode credentials or targets

Contributing

See CONTRIBUTING.md in the repository for guidelines on:

• Adding new phases
• Extending router logic
• Creating custom packs
• Submitting bug reports

License

MIT License - See LICENSE file for details.