Adversarial Code Generator

Generate secure, verified code using adversarial AI testing directly in VS Code.

⚠️ Requirements

This extension requires a local backend server to function. The backend performs adversarial security verification on generated code.

Prerequisites

1. Python 3.9 or higher
2. Git
3. API key for at least one LLM provider:
   • OpenRouter (recommended)
   • OpenAI
   • Claude/Anthropic
   • Google Gemini

🚀 Quick Setup

Step 1: Clone and Start Backend Server

# Clone the repository
git clone https://github.com/a7madiv/adversarial-verification-framework
cd adversarial-verification-framework/backend

# Install dependencies
pip install -r requirements.txt

# Start the server
python api/server.py

The server will start on http://localhost:5001

Step 2: Configure VS Code Extension

1. Install this extension from the marketplace
2. Open VS Code Settings (Ctrl+,)
3. Search for "Secure LLM"
4. Configure:
   • Server URL: http://localhost:5001 (default)
   • AI Provider: Choose your provider (openrouter, openai, claude, gemini)
   • API Key: Enter your API key for the chosen provider

Step 3: Start Generating Secure Code!

• Click the "Adversarial" button in the status bar, or
• Press Ctrl+Shift+L (Windows/Linux) or Cmd+Shift+L (Mac)
• Enter your code requirement
• Get secure, verified code with security reports!

✨ Features

• Adversarial Verification: Code is tested against security attacks before delivery
• Multi-Provider Support: Works with OpenAI, Claude, OpenRouter, and Gemini
• Security Reports: Every generation includes detailed security analysis
• Multiple Verification Layers: Static analysis, fuzzing, and dynamic testing
• Confidence Scoring: Know how secure your generated code is

📝 Example Requirements

Try these prompts:

• "Write a function to validate email format using regex"
• "Create a secure password validation function"
• "Generate a function for safe file upload handling"
• "Write a function to sanitize SQL query inputs"

🔒 How It Works

1. Blue Agent: Generates code using your chosen LLM
2. Static Gate: Analyzes code for security issues (Bandit, Semgrep, Pylint)
3. Red Agent: Generates adversarial test cases and exploits
4. Dynamic Gate: Executes code in sandboxed environment
5. Adjudicator: Makes final security verdict (PASS/WARN/FAIL)

📊 What You Get

For each code generation:

• Main Code File: Secure, verified implementation
• Security Report: Detailed analysis including:
  • Verification status (PASS/WARN/FAIL)
  • Confidence score
  • Security violations found
  • Evidence and recommendations

⚙️ Configuration Options

🛠️ Troubleshooting

"Cannot connect to server"

• Make sure the backend server is running: python api/server.py
• Check the server URL in settings matches: http://localhost:5001

"API key not found"

• Configure your API key in VS Code settings
• Or set environment variable in backend .env file

"Request timeout"

• Increase timeout in settings (default: 120 seconds)
• Complex requirements may take longer

📚 Documentation

• Full Documentation
• Architecture Guide
• Security Methodology

🤝 Contributing

Contributions welcome! See the main repository for details.

📄 License

MIT License - see LICENSE for details.

🔗 Links

• GitHub Repository
• Report Issues
• Marketplace

Note: This extension is designed for developers who want to generate secure code with confidence. The adversarial verification process ensures code is tested against real security attacks before you use it.