Inducing Programmatic Skills for Agentic Tasks

2025-04-09

AI Summary

To address the poor generalization and adaptability of agents in digital tasks such as web navigation, this paper proposes the Agent Skill Induction (ASI) framework. ASI represents skills as verifiable, programmatic abstractions, enabling autonomous online skill induction, feedback-driven execution validation, cross-site transfer, and dynamic skill updating under environmental changes. The method integrates program synthesis, action abstraction and composition, online skill induction, and execution-result-based verification. Evaluated on the WebArena benchmark, ASI achieves a 23.5% absolute improvement in task success rate over static baselines, while reducing execution steps by 10.7–15.3%, demonstrating significantly enhanced generalization and environmental robustness. Its core contribution is the first realization of a closed-loop pipeline for online skill generation, verification, and reuse—establishing a novel paradigm for agents to continuously evolve their programmable capabilities through interaction.

Technology Category

• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

To succeed in common digital tasks such as web navigation, agents must carry out a variety of specialized tasks such as searching for products or planning a travel route. To tackle these tasks, agents can bootstrap themselves by learning task-specific skills online through interaction with the web environment. In this work, we demonstrate that programs are an effective representation for skills. We propose agent skill induction (ASI), which allows agents to adapt themselves by inducing, verifying, and utilizing program-based skills on the fly. We start with an evaluation on the WebArena agent benchmark and show that ASI outperforms the static baseline agent and its text-skill counterpart by 23.5% and 11.3% in success rate, mainly thanks to the programmatic verification guarantee during the induction phase. ASI also improves efficiency by reducing 10.7-15.3% of the steps over baselines, by composing primitive actions (e.g., click) into higher-level skills (e.g., search product). We then highlight the efficacy of ASI in remaining efficient and accurate under scaled-up web activities. Finally, we examine the generalizability of induced skills when transferring between websites, and find that ASI can effectively reuse common skills, while also updating incompatible skills to versatile website changes.

Problem

• Inducing program-based skills for web agent tasks
• Improving agent efficiency and accuracy in web navigation
• Enhancing skill transferability across different websites

Innovation

• Program-based skills for agent adaptation
• Dynamic skill induction and verification
• Composing primitive actions into higher-level skills