SearchSkill: Teaching LLMs to Use Search Tools with Evolving Skill Banks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hu, Jinchao, Zhong, Meizhi, Chen, Kehai, Zhang, Min
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909042836766720
author Hu, Jinchao
Zhong, Meizhi
Chen, Kehai
Zhang, Min
author_facet Hu, Jinchao
Zhong, Meizhi
Chen, Kehai
Zhang, Min
contents Teaching language models to use search tools is not only a question of whether they search, but also of whether they issue good queries. This is especially important in open-domain question answering, where broad or copied queries often waste retrieval budget and derail later reasoning. We propose \Ours, a framework that makes query planning explicit through reusable search skills. At each step, the model first selects a skill, then generates a search or answer action conditioned on the selected skill card. The skill inventory itself is not fixed: SearchSkill maintains an evolving SkillBank, expands or refines it from recurrent failure patterns, and reconstructs affected trajectories before supervised training. The resulting two-stage SFT recipe aligns training with the inference-time protocol of skill selection followed by skill-grounded execution. Across open-source and closed-source models, SearchSkill improves exact match on knowledge-intensive QA benchmarks and yields better retrieval behavior, including fewer copied first queries, more atomic hop-focused queries, and more correct answers within a small search budget. These results suggest that explicit skill-conditioned query planning is a lightweight alternative to treating search as an undifferentiated action.
format Preprint
id arxiv_https___arxiv_org_abs_2605_09038
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SearchSkill: Teaching LLMs to Use Search Tools with Evolving Skill Banks
Hu, Jinchao
Zhong, Meizhi
Chen, Kehai
Zhang, Min
Artificial Intelligence
Teaching language models to use search tools is not only a question of whether they search, but also of whether they issue good queries. This is especially important in open-domain question answering, where broad or copied queries often waste retrieval budget and derail later reasoning. We propose \Ours, a framework that makes query planning explicit through reusable search skills. At each step, the model first selects a skill, then generates a search or answer action conditioned on the selected skill card. The skill inventory itself is not fixed: SearchSkill maintains an evolving SkillBank, expands or refines it from recurrent failure patterns, and reconstructs affected trajectories before supervised training. The resulting two-stage SFT recipe aligns training with the inference-time protocol of skill selection followed by skill-grounded execution. Across open-source and closed-source models, SearchSkill improves exact match on knowledge-intensive QA benchmarks and yields better retrieval behavior, including fewer copied first queries, more atomic hop-focused queries, and more correct answers within a small search budget. These results suggest that explicit skill-conditioned query planning is a lightweight alternative to treating search as an undifferentiated action.
title SearchSkill: Teaching LLMs to Use Search Tools with Evolving Skill Banks
topic Artificial Intelligence
url https://arxiv.org/abs/2605.09038