Results of the summarizing process for the arXiv paper: 2504.07079v1

In complex environments, humans have evolved sophisticated self-improvement mechanisms through environment exploration, hierarchical abstraction of experiences into reusable skills, and collaborative construction of a skill repertoire. However, autonomous web agents still struggle with procedural knowledge abstraction, refining skills, and skill composition. To address this challenge, SkillWeaver is introduced as a skill-centric framework enabling agents to autonomously synthesize reusable skills as APIs. The framework consists of three stages: 1. Skill Proposal: The agent identifies novel skills based on observations and available APIs in the skill library. 2. Skill Synthesis: Successful trajectories generated by executing proposed skills are used to synthesize APIs. 3. Skill Honing: Synthesized APIs undergo testing with automatically generated test cases for robustness. Through iterative exploration and practice on new websites, the agent builds a library of lightweight, plug-and-play APIs that enhance its capabilities significantly. Experiments on WebArena and real-world websites demonstrate the efficacy of SkillWeaver, with relative success rate improvements of 31.8% and 39.8%, respectively. Stronger agents can also enhance weaker ones through transferable skills synthesized by SkillWeaver, resulting in improvements of up to 54.3% on WebArena. By honing diverse website interactions into APIs that can be seamlessly shared among web agents, SkillWeaver showcases the effectiveness of autonomous self-improvement in navigating complex online environments. This framework enables agents to build conceptual maps of website environments, accumulate procedural knowledge as reusable skills, compose simple skills into complex routines, and enhance decision-making processes through learned skills without the need for extensive training data or external supervision.

• Humans have evolved self-improvement mechanisms through environment exploration, hierarchical abstraction of experiences, and collaborative construction of skill repertoires
• Autonomous web agents struggle with procedural knowledge abstraction, refining skills, and skill composition
• SkillWeaver framework enables agents to synthesize reusable skills as APIs through three stages:
• Skill Proposal: Identifying novel skills based on observations and available APIs
• Skill Synthesis: Generating successful trajectories from proposed skills to synthesize APIs
• Skill Honing: Testing synthesized APIs with automatically generated test cases for robustness
• Experiments show significant success rate improvements on WebArena and real-world websites using SkillWeaver
• Stronger agents can enhance weaker ones through transferable skills synthesized by SkillWeaver
• SkillWeaver enables autonomous self-improvement in navigating complex online environments by building conceptual maps, accumulating procedural knowledge as reusable skills, composing simple skills into complex routines, and enhancing decision-making processes without extensive training data or external supervision

Introduction

The internet has become an integral part of our daily lives, with millions of websites offering a vast array of information and services. Navigating through this complex online environment can be challenging for humans, but even more so for autonomous web agents. These agents are computer programs designed to perform specific tasks on the internet without human intervention.

In recent years, there has been significant progress in developing autonomous web agents that can perform various tasks such as data extraction, form filling, and web scraping. However, these agents still struggle with procedural knowledge abstraction, refining skills, and skill composition. This is where SkillWeaver comes in – a skill-centric framework that enables autonomous web agents to autonomously synthesize reusable skills as APIs.

The Evolution of Self-Improvement Mechanisms

Humans have evolved sophisticated self-improvement mechanisms through environment exploration, hierarchical abstraction of experiences into reusable skills, and collaborative construction of a skill repertoire. This allows us to adapt to new situations and challenges quickly. Similarly, SkillWeaver aims to provide autonomous web agents with the ability to improve their capabilities through iterative exploration and practice on new websites.

Environment Exploration

One key aspect of self-improvement is exploring one's environment. Humans do this by trying out different approaches or techniques until they find one that works best for them. In the same way, SkillWeaver allows autonomous web agents to explore new websites by identifying novel skills based on observations and available APIs in the skill library.

Hierarchical Abstraction

Another crucial element in human self-improvement is hierarchical abstraction – breaking down complex tasks into smaller subtasks or skills that can be reused in different situations. For example, learning how to ride a bike involves mastering several smaller skills like balancing and pedaling. Similarly, SkillWeaver enables autonomous web agents to break down website interactions into smaller, reusable skills that can be synthesized into APIs.

Collaborative Construction

Humans also learn from each other through collaboration and sharing of knowledge. This collaborative construction of a skill repertoire allows us to build upon the skills of others and enhance our own capabilities. SkillWeaver takes this concept a step further by allowing autonomous web agents to share their synthesized skills with each other, enhancing their overall performance.

The Three Stages of SkillWeaver

SkillWeaver consists of three stages – Skill Proposal, Skill Synthesis, and Skill Honing. These stages work together to enable autonomous web agents to improve their capabilities in navigating complex online environments.

Skill Proposal

In the first stage, the agent identifies novel skills based on observations and available APIs in the skill library. This is similar to how humans explore new environments by trying out different approaches until they find one that works best for them.

Skill Synthesis

Once a novel skill has been identified, successful trajectories generated by executing proposed skills are used to synthesize APIs. This process involves breaking down website interactions into smaller reusable skills that can be composed into more complex routines.

Skill Honing

The final stage involves testing the synthesized APIs with automatically generated test cases for robustness. This ensures that the newly created API is reliable and can perform its intended task effectively. Through iterative exploration and practice on new websites, autonomous web agents using SkillWeaver can build a library of lightweight, plug-and-play APIs that significantly enhance their capabilities.

Experiments and Results

To demonstrate the effectiveness of SkillWeaver, experiments were conducted on WebArena (a simulated environment) as well as real-world websites. The results showed significant improvements in success rates – 31.8% on WebArena and 39.8% on real-world websites.

Furthermore, SkillWeaver also allows stronger agents to enhance weaker ones through transferable skills synthesized by the framework. This resulted in improvements of up to 54.3% on WebArena, showcasing the collaborative aspect of self-improvement in navigating complex online environments.

Conclusion

In conclusion, SkillWeaver is a skill-centric framework that enables autonomous web agents to autonomously synthesize reusable skills as APIs. By honing diverse website interactions into APIs that can be seamlessly shared among web agents, SkillWeaver showcases the effectiveness of autonomous self-improvement in navigating complex online environments.

This framework not only allows for iterative exploration and practice on new websites but also enables agents to build conceptual maps of website environments, accumulate procedural knowledge as reusable skills, compose simple skills into complex routines, and enhance decision-making processes through learned skills without the need for extensive training data or external supervision. With further development and implementation, SkillWeaver has the potential to greatly improve the capabilities of autonomous web agents and revolutionize their role in our increasingly digital world.