dreadnode/AIRTBench-Code

Implements an autonomous red teaming agent using a modular harness architecture that executes Python code within isolated Docker containers to solve AI/ML CTF challenges. The agent integrates with the Dreadnode Strikes platform and uses the Rigging framework, receiving challenge notebooks and iteratively attempting exploits through a Jupyter kernel feedback loop with configurable step limits and timeout controls. Supports filtering for LLM-based challenges and provides standardized evaluation metrics for measuring adversarial capabilities across different language models.