AI News

Web agents frequently revisit the same websites, but standard evaluations often discard procedures acquired during earlier successful interactions. EconSkills introduces a skill library and evaluation framework that distills verified EconWebArena trajectories into parameterized standard operating procedures for retrieving live economic data. Each skill captures its scope, navigation procedure, site-specific guidance, verification checks, and recovery steps, while replacing source-specific values with placeholders. The framework separates transfer of a known relevant procedure to a held-out task from the harder problem of selecting useful skills from a library. In controlled transfer experiments, matched skills improve success over no-skill prompting and use fewer steps on paired successful tasks. Abstracted procedures perform substantially better than replaying raw trajectories. At library scale, retrieval is competitive with the no-skill baseline overall and performs best on tasks directly covered by the library, but approximate matches on uncovered tasks offset those gains. Browser trajectories also show when procedural guidance reduces portal-specific navigation and when semantic verification is still required. The results support coverage-aware skill selection and context delivery as design targets for reusable web-agent procedures.