Inducing Programmatic Skills for Agentic Tasks

TL;DR

This work tackles the challenge of enabling web navigation agents to autonomously acquire and use task-oriented skills online. It introduces ASI, which induces executable programmatic skills, verifies them via execution, and embeds them directly into the agent's action space, enabling reusable, verifiable workflows. On WebArena, ASI achieves substantial gains in both success rate (up to 23.5% over static baselines and 11.3% over text-based adaptive agents) and planning efficiency (10.7–15.3% fewer steps), with pronounced benefits on longer-horizon tasks and cross-website generalization. The results demonstrate that programmatic skills offer verifiability, composability, and effective transfer, supporting scalable, adaptive web navigation across varied domains and real-world sites.

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.

Figures (4)

• Figure 1: Online adaptive agent that induces and reuses programmatic skills as actions (bottom), as opposed to adding textual skills in memory (top).
• Figure 2: Inducing programmatic skills and rewriting the trajectory from an episode.
• Figure 3: Example scaled-up task of updating multiple addresses on shopping website.
• Figure 4: ASI can generalize the search_product skill but face incompatibility when sorting items.