---
title: "From Local Skills to Long-Horizon Web Tasks: Progressive Skill Exploration for LLM Web Agents"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.03116v1"
permalink: "https://feed.astralor.com/items/itm_50ef6a01aaae2d244392f54c"
published: "2026-11-02T13:32:59.000Z"
score: 59
channel: "finance"
tags: ["LLM","AI Agent","RAG","From Local Skills","Long-Horizon Web Tasks","Progressive Skill Exploration","LLM Web Agents","Large"]
---

# From Local Skills to Long-Horizon Web Tasks: Progressive Skill Exploration for LLM Web Agents

## At a glance

Large language models (LLMs) are driving autonomous agents toward real-world environments, with the Web emerging as one of the most representative interaction scenarios. However, training Web agents with naive explorati…

## Key points

- Large language models (LLMs) are driving autonomous agents toward real-world environments, with the Web emerging as one of the most representative interaction scenarios.
- However, training Web agents with naive exploration remains challenging: successful trajectories are rare, and the sparse terminal rewards make it difficult for agents to obtain useful feedback from failed trajectories.
- Inspired by skill acquisition theory, we propose Skill-Composed Progressive Exploration (SCOPE), a short-to-long learning framework for long-horizon Web agents that decomposes complex task exploration into a gradual exp…

## Full text

Large language models (LLMs) are driving autonomous agents toward real-world environments, with the Web emerging as one of the most representative interaction scenarios. However, training Web agents with naive exploration remains challenging: successful trajectories are rare, and the sparse terminal rewards make it difficult for agents to obtain useful feedback from failed trajectories. Inspired by skill acquisition theory, we propose Skill-Composed Progressive Exploration (SCOPE), a short-to-long learning framework for long-horizon Web agents that decomposes complex task exploration into a gradual expansion process from local skills to complete task-solving capability. Specifically, SCOPE iteratively explores around local skills: Skill Discovery extracts reusable local skills and failure experiences from historical rollouts; Skill Orchestration composes discovered skills into high-level skill chains; and Skill Expansion leverages failure experiences aligned with the current skill chain to guide the agent toward exploring new subskills. Through this process, SCOPE progressively constructs task-completing skill chains across rollouts and internalizes them into the policy parameters through agentic on-policy distillation. Extensive experiments on WebArena, WebVoyager, and WebShop show that SCOPE consistently improves exploration dynamics and achieves 3.9% to 18.6% relative gains over the strongest baselines in each setting.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_50ef6a01aaae2d244392f54c