WebXSkill: Skill Learning for Autonomous Web Agents

2025-10-10 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

WebXSkill is a novel framework designed to enhance autonomous web agents powered by large language models (LLMs) by addressing the "grounding gap" in existing skill formulations. It introduces "executable skills" that combine parameterized action programs with step-level natural language guidance, allowing for both direct execution and agent-driven adaptation. The framework operates in three stages: skill extraction from synthetic agent trajectories, skill organization into a URL-based graph for context-aware retrieval, and skill deployment in two modes. The "grounded mode" enables fully automated multi-step execution, while the "guided mode" provides step-by-step instructions for the agent's native planning. Evaluated on WebArena and WebVoyager benchmarks, WebXSkill improved task success rates by up to 9.8 and 12.9 points over baselines, respectively, demonstrating the effectiveness of its dual-nature skills.

Key takeaway

Research Scientists developing autonomous web agents should consider implementing WebXSkill's dual-mode executable skills to improve task success and adaptability. If your LLM is robust, prioritize grounded mode for efficiency; for less capable models, guided mode offers better error recovery and adaptation, especially for cross-environment skill transfer. This approach reduces re-planning and enhances procedural knowledge reuse.

Key insights

WebXSkill bridges the "grounding gap" for web agents by combining executable actions with natural language guidance.

Principles

• Skills should be both executable and interpretable.
• Context-aware retrieval improves skill adoption.
• Deployment mode should adapt to model capability.

Method

WebXSkill extracts parameterized skills from synthetic trajectories, organizes them into a URL-based graph, and deploys them in either grounded (automated) or guided (step-by-step) modes.

In practice

• Use synthetic trajectories for skill extraction.
• Organize skills in a URL-based graph for retrieval.
• Implement dual deployment modes for flexibility.

Topics

• Autonomous Web Agents
• LLM-powered Skill Learning
• Executable Skills
• Grounding Gap
• Skill Graph Organization

Code references

• aiming-lab/WebXSkill
• browser-use/browser-use

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• WebXSkill: Skill Learning for Autonomous Web Agents — WebXSkill bridges the skill grounding gap for LLM-powered web agents using executable skills with natural language guidance.
• Beyond Domains: Reusing Web Skills via Transferable Interaction Patterns — Web agent skill reuse is significantly improved by matching layout structure rather than specific element references.
• Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward — Agent skills modularize LLM capabilities, extending functionality dynamically while introducing significant security challenges.
• RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources — Distilling multimodal human-created resources into executable, reusable skills significantly enhances software agent performance.
• Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills — Trace2Skill distills broad agent experience into transferable, declarative skills via parallel analysis and inductive consolidation.
• SkillCAT: Contrastive Assessment and Topology-Aware Skill Self-Evolution for LLM Agents — SkillCAT enables LLM agents to self-evolve skills by contrastive assessment and topology-aware execution, boosting performance without training.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.