When to End, Which to Invoke Next: Transition-Aware Tokenized Procedural Memory for LLMs

$\text{One-Token Procedural Memory}$ ($\text{TokMem}$) has recently emerged as a parameter-efficient paradigm for storing and reusing task-solving procedures in large language models (LLMs), where each procedure is encoded as a compact memory token. This paper investigates a more challenging scenario in which $\text{a LLM must reliably orchestrate a sequence of consecutive procedure recalls within a single generation}$. We identify a $\text{fundamental limitation}$ of TokMem-style procedure chaining: while memory tokens provide explicit entry points to individual procedures, the transitions between procedures remain implicit. This overlooked handoff mechanism causes next-procedure selection to become increasingly fragile as procedural chains grow longer. To address this issue, we propose $\text{Transition-Aware Procedural Memory}$ ($\text{TapMem}$), a lightweight framework that explicitly models inter-procedure handoffs. The key insight behind TapMem is to formulate each procedure transition as a boundary-conditioned selection problem. Since the next procedure should be selected only after the current procedure-guided span ends, TapMem first introduces $\text{E}$xplicit $\text{T}$ermination $\text{S}$ignal, to make this completion boundary observable. Building on the resulting boundary representation, TapMem further incorporates a $\text{T}$ransition-$\text{C}$onditioned $\text{R}$outing $\text{A}$dapter to generate a context-dependent correction over candidate memory-token logits. By intervening only at transition positions and only over memory-token logits, TapMem preserves the original full-vocabulary autoregressive generation process while improving the sparse procedural decisions required for reliable multi-step orchestration. Extensive experiments across diverse procedure-recall settings demonstrate that TapMem achieves promising performance.