Inducing Reasoning Primitives from Agent Traces

Researchers introduced Reasoning Primitive Induction, a single-pass method that mines successful ReAct agent traces to extract recurrent reasoning moves into a compact library of typed pseudo-tools. The induced libraries outperformed the original agents that generated their traces, achieving gains of up to +44 percentage points on complex reasoning benchmarks. Across five subtasks spanning narrative deduction and constraint-satisfaction planning, the method matched or surpassed expert-authored decompositions while operating at lower inference cost.

arXiv:2606.02994v1 Announce Type: new 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 - 74 , +30pp on MuSR team allocation 38 - 68 , and +22pp on NatPlan meeting planning 7 - 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.