Research questionHow can LLM agents reuse execution traces without losing temporal and outcome-dependent behavior?LLM agents often retrieve or summarize past trajectories too shallowly, losing dependencies across steps and the conditions associated with success or failure. Effective reuse must preserve how behaviors unfold over time while distinguishing outcome-relevant patterns from unreliable shortcuts.