From Memory to Guide: Spatio-Temporal Composer for Procedural Coding Memory

2026-10-04

AI Summary

This study addresses the disconnect between textual readability and execution reliability caused by retrieved knowledge injection in memory-augmented agents. To overcome this, we propose the Spatiotemporal Combiner paradigm, which transforms static procedural memory into dynamic runtime guidance, enabling active strategy adaptation during inference through spatial alignment and temporal orchestration. Notably, this work introduces a pioneering mechanism that dynamically aligns historical knowledge with current environmental constraints without parameter updates, while precisely controlling the knowledge lifecycle via local goal setting and behavioral guardrails. Evaluated on the EngramBench benchmark, our approach improves task success rates by 7.2% and reduces token consumption by 32.2%, effectively mitigating context mismatch issues.

Abstract

Memory-augmented agents typically integrate procedural knowledge by injecting retrieved skills directly into text prompts. This approach dangerously equates readable text with reliable execution. To bridge this gap, we introduce From Memory to Guide, a novel paradigm that transitions procedural memory from passive text delivery to active, inference-time policy adaptation. We instantiate this paradigm through the Spatio-Temporal Composer, an active policy compiler that explicitly manages exactly how and when retrieved knowledge should be applied. Rather than treating skills as plug-and-play modules, Composer dynamically aligns historical knowledge with current environmental constraints (spatial adaptation) and precisely dictates its applicable lifecycle (temporal orchestration). It actively transforms static memories into strictly bounded Runtime Guides---equipping the agent with localized objectives and behavioral guardrails without requiring a single parameter update. Extensive evaluations on 13 demanding, long-horizon software engineering tasks in EngramBench demonstrate the clear advantages of this architecture. Composer not only robustly prevents context mismatch but drives an absolute pass-rate increase of 7.2 percentage points on the most complex tasks, while simultaneously slashing the main agent's token usage by 32.2%.

Problem

• memory-augmented agents
• procedural memory
• context mismatch
• policy adaptation
• software engineering tasks

Innovation

• Spatio-Temporal Composer
• Memory-augmented Agents
• Procedural Memory
• Inference-time Policy Adaptation
• Runtime Guides