AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

One Skill, Many Websites: Learning Generalizable Skills Through Polymorphic Abstraction

Simon Yu, Gang Li, Weiyan Shi, Peng Qi

ICLR 2026 · Multi-agent systems and games › LLM agents · approach: LLM multi-agent collaboration and safety

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes PolySkill, a framework that has LLM agents learn reusable web skills via polymorphic abstraction so they generalize across websites rather than overfitting to one site.

Problem

Skill learning for LLM web agents tends to produce skills over-specialized to a single website, which fail to transfer to other sites.

Approach

Abstracts agent-learned skills polymorphically so that a single skill can apply across multiple websites, rather than being tied to one site's specifics. Only the abstract portion is available; details beyond it are unknown.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add polymorphic abstract-class skill structureASI-style skill induction (static) · Mind2Web Cross-tasksuccess rate: 52.3 -> 55.4Improved
add polymorphic structureskill induction with online update · Mind2Web Cross-tasksuccess rate: 59.4 -> 63.2Improved
add polymorphic structurestatic skill induction · Mind2Web Cross-Domainsuccess rate: 57.3 -> 60.1 vs baseline 62.3No clear effect
static skill induction without polymorphism versus no skillsASI on GPT-4.1 · Mind2Websuccess rate: 53.8 -> 52.3 (cross-task)Hurt
add polymorphic structureskill library size · Mind2Web Cross-tasknumber of skills: 66 -> 47 (with update)Improved
add online continual update (ablation)PolySkill · Mind2Web Cross-tasksuccess rate: 55.4 -> 63.2Improved
train on a single website only (ablation)skill learner · Target / other sitesskill reusability: below 4% on other sitesHurt
use self-guided exploration instead of fixed sequential curriculum (ablation)PolySkill · WebArena Shoppingsuccess rate: 42.1 -> 43.1Improved
use self-guided exploration instead of SkillWeaver curriculum (ablation)skill induction · WebArena Shoppingsuccess rate: 39.8 -> 43.1Improved
specialize on new sites after learning on WebArenaASI vs PolySkill · WebArena Shoppingsuccess rate: +4.9% for PolySkill over ASIImproved
reordering sequential curriculum (ablation)PolySkill · WebArena Shoppingsuccess rate: 38.2 to 42.1 across ordersMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy