Scalable Behaviour Cloning on Browser Using via Skill Distillation

2026-06-30

AI Summary

This work addresses the limited decision-making capabilities of existing browser agents in information-scarce scenarios and their difficulty in distilling reusable skills from vast amounts of human browsing behavior. To overcome these challenges, the authors propose a “skill distillation” approach that transforms raw interaction trajectories into compact natural language skill descriptions. These distilled skills are organized into a structured, retrievable, and composable skill graph that supports continuous evolution, replacing conventional task-specific behavioral cloning paradigms. By effectively capturing implicit priors embedded in human interactions, the method enables structured representation and efficient reuse of browsing skills. Experimental results demonstrate that the proposed framework significantly enhances both decision-making competence and generalization performance of agents in complex web navigation tasks.

Abstract

Internet users collectively perform an enormous range of skilled work through web browsers, from software development and document editing to search, forms, and enterprise workflows, making human browsing a highly scalable but under-exploited source of reusable browser skills. We argue that the bottleneck for browser agents is decision-making under incomplete information rather than low-level operation, and that the priors agents lack are already implicit in human interaction traces. We therefore study scalable behavior cloning for browser agents via skill distillation, converting user interaction trajectories into compact natural-language skills that agents can read, retrieve, reuse, and compose directly. We further organize the distilled skills into a skill graph so that growth proceeds through consolidation rather than unbounded accumulation. This suggests that the scalability of browser agents may come less from manually designed tasks and more from the collective skills already expressed by internet users. Our project is available at: https://lab.einsia.ai/browserbc/.

Problem

• browser agents
• behavior cloning
• skill distillation
• human interaction traces
• scalability

Innovation

• behavior cloning
• skill distillation
• browser agents
• skill graph
• human interaction traces