AutoVuln-Builder: LLM-Powered CyberRange Generator

AutoVuln-Builder is an autonomous DevSecOps agent that translates natural language requests into ready-to-deploy, intentionally vulnerable Infrastructure as Code (IaC) for cybersecurity training and CyberRange environments.

Architecture & Guardrails

This project demonstrates how to tame a Large Language Model (LLM) for precise infrastructure generation by decoupling the "decision making" from the "code writing".

1. The Brain (LLM & LangChain): Analyzes the user's scenario request and selects the appropriate services, ports, and vulnerabilities.
2. The Guardrail (Pydantic): Enforces strict mathematical and logical rules (e.g., valid ports, allowed vulnerability types, database dependencies) to prevent AI hallucinations.
3. The Factory (Jinja2): Takes the validated AI output and injects it into rigid, syntactically perfect templates to generate docker-compose.yml and .Dockerfile files.

Features

• Natural Language to Infrastructure (Prompt-to-IaC).
• Autonomous database provisioning if an application requires it.
• Built-in conditional logic (e.g., dynamically installing mysqli only for PHP containers).
• 100% syntactically correct YAML and Dockerfile generation.

Prerequisites

• Python 3.10+
• Docker & Docker Compose
• OpenAI API Key

Installation

1. Clone the repository:

   git clone https://github.com/Augustin-Br/autovuln_builder.git
   cd autovuln_builder

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set up your API Key: Rename .env.example to .env and add your OpenAI API key:

   OPENAI_API_KEY=sk-...

Usage

1. Generate the Scenario: Run the orchestrator script. It will call the AI and generate the infrastructure files in the output_scenario/ folder.

   python main.py

2. Deploy the CyberRange:

   cd output_scenario
   sudo docker compose up --build

3. Test the Vulnerability (Example: SQLi): Open a new terminal and send an encoded payload to the generated web service:

   curl -v -G "http://localhost:8080/index.php" --data-urlencode "user=' OR 1=1 -- "

Watch the Docker logs to confirm the payload successfully traversed the PHP backend and hit the database layer!