Search papers, labs, and topics across Lattice.
This paper introduces $S^3$, a multi-stage defense framework designed to enhance the safety of Large Language Model (LLM) agents by integrating stage-specific safety skills into their workflows. By addressing the limitations of existing safety methods that only protect isolated stages, $S^3$ enables comprehensive risk detection and mitigation across memory, planning, and tool execution stages. Experimental results indicate that $S^3$ significantly outperforms state-of-the-art safety solutions in both effectiveness and utility preservation, highlighting the potential of a unified safety approach for agent systems.
Stage-specific safety skills can transform LLM agent safety by providing a scalable and composable framework that outperforms existing methods.
Large Language Model (LLM) agents rely on multi-stage agentic workflows, with stages such as memory, planning, and tool execution, to accomplish complex tasks. However, risks may emerge at different stages, propagate across steps, and become difficult to detect and mitigate. Existing safety methods protect only isolated stages and are difficult to integrate, leaving agents without comprehensive protection throughout the workflow. To address these limitations, we introduce Stage-Specific Safety Skills, a unified abstraction that represents heterogeneous safety designs as reusable and composable components with explicit stage semantics. We further develop an automated transformation pipeline that converts existing safety designs into reusable safety skills and establish a community-driven safety skill library. Building on this abstraction, we propose $S^3$, a multi-stage defense framework in which a guard agent orchestrates stage-specific safety skills for risk detection and mitigation throughout the agentic workflow. We also construct the Multi-Stage Risk Benchmark (MSRB) to evaluate representative risks across workflow stages. Experimental results show that $S^3$ consistently outperforms representative state-of-the-art baselines in both safety effectiveness and utility preservation. These results demonstrate the potential of stage-specific safety skills as a scalable and composable foundation for building resilient and trustworthy agent systems.
