{"ID":3006154,"CreatedAt":"2026-06-03T03:09:48.883664427Z","UpdatedAt":"2026-06-04T19:14:31.964469513Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2606.02994","arxiv_id":"2606.02994","title":"Inducing Reasoning Primitives from Agent Traces","abstract":"ReAct-style LLM agents often rediscover the same reasoning routines across problems, yet leave those routines trapped in transient scratchpads. We introduce Reasoning Primitive Induction, a single-pass method that mines successful ReAct traces, clusters recurrent reasoning moves, and converts the most frequent moves into a compact library of typed pseudo-tools. Each pseudo-tool is specified by a natural-language docstring interpreted by an LLM at invocation time, and a standard ReAct loop composes these primitives at test time. The central result is that induced libraries outperform the very agent that generated their traces: by +44pp on RuleArena NBA (30 -\u003e 74), +30pp on MuSR team allocation (38 -\u003e 68), and +22pp on NatPlan meeting planning (7 -\u003e 29). Across five comparable subtasks spanning narrative deduction, rule application, and constraint-satisfaction planning, a single fixed configuration improves over zero-shot Chain-of-Thought on every subtask, matches or surpasses expert-authored decompositions, and outperforms AWM at lower average inference cost.","short_abstract":"ReAct-style LLM agents often rediscover the same reasoning routines across problems, yet leave those routines trapped in transient scratchpads. We introduce Reasoning Primitive Induction, a single-pass method that mines successful ReAct traces, clusters recurrent reasoning moves, and converts the most frequent moves in...","url_abs":"https://arxiv.org/abs/2606.02994","url_pdf":"https://arxiv.org/pdf/2606.02994v1","authors":"[\"Zhihan Lei\",\"Jiarui Yan\",\"Joshua Momo\",\"William W. Cohen\"]","published":"2026-06-02T01:11:15Z","proceeding":"cs.AI","tasks":"[\"cs.AI\",\"cs.CL\"]","methods":"[\"Large Language Model\"]","has_code":false}