Gravelroot automatically generates attack graphs that represent potential vulnerability propagation paths, helping developers identify high-risk method calls driven by LLM outputs, understand how security issues are triggered, and evaluate the overall safety of Agents under complex interaction scenarios.

The repository of TrafficLLM, a universal LLM adaptation framework to learn robust traffic representation for all open-sourced LLM in real-world scenarios and enhance the generalization across diverse traffic analysis tasks.