The codebase for our paper Agent-SafetyBench: Evaluating the Safety of LLM Agents . Agent-SafetyBench is a comprehensive agent safety evaluation benchmark that introduces a diverse array of novel environments that are previously unexplored, and offers broader and more systematic coverage of…