S2Bench is a specialized dataset designed to evaluate the security and robustness of Large Language Model (LLM) agents in complex dynamic environments. It is a core component of the Spider-Sense framework, covering the full lifecycle from planning to execution.
Stage Action: Evaluates the compliance and safety of agent actions.
Stage Observation: Evaluates agent… See the full description on the dataset page:
https://huggingface.co/datasets/aifinlab/S2Bench.
