Agent memory portability test reveals hidden lock-in risks

A developer proposes a round-trip test to check if agent memory systems can survive framework changes, model deprecations, or multi-agent sharing. The test exposes five design decisions—salience, supersession, export tiers, embedding traps, and fleet coordination—that most implementations ignore. The post argues memory is fundamentally a write problem, not a retrieval one, and offers a checklist for portable memory.