FocusAgent: Simple Yet Effective Ways of Trimming the Large Context of Web Agents
This addresses efficiency and security issues for web agents powered by LLMs, though it is incremental as it builds on existing pruning strategies with a novel retrieval-based method.
The paper tackled the problem of large context sizes in web agents by introducing FocusAgent, which uses a lightweight LLM retriever to prune irrelevant content from web page observations, reducing observation size by over 50% while matching baseline performance on benchmarks and significantly lowering vulnerability to prompt-injection attacks.
Web agents powered by large language models (LLMs) must process lengthy web page observations to complete user goals; these pages often exceed tens of thousands of tokens. This saturates context limits and increases computational cost processing; moreover, processing full pages exposes agents to security risks such as prompt injection. Existing pruning strategies either discard relevant content or retain irrelevant context, leading to suboptimal action prediction. We introduce FocusAgent, a simple yet effective approach that leverages a lightweight LLM retriever to extract the most relevant lines from accessibility tree (AxTree) observations, guided by task goals. By pruning noisy and irrelevant content, FocusAgent enables efficient reasoning while reducing vulnerability to injection attacks. Experiments on WorkArena and WebArena benchmarks show that FocusAgent matches the performance of strong baselines, while reducing observation size by over 50%. Furthermore, a variant of FocusAgent significantly reduces the success rate of prompt-injection attacks, including banner and pop-up attacks, while maintaining task success performance in attack-free settings. Our results highlight that targeted LLM-based retrieval is a practical and robust strategy for building web agents that are efficient, effective, and secure.