Conversation Memory

This is a multi-tier memory architecture for LLM conversations that splits storage into short-term, long-term, and entity-based layers. You'd reach for this when building chatbots or assistants that need to recall user context across sessions without bloating every prompt with irrelevant history. The skill emphasizes retrieval strategy over just dumping everything into storage, which is the right instinct. Watch out for unbounded growth and cross-user memory leaks, both flagged as high severity issues. Pairs naturally with context window management and RAG patterns. The source material is somewhat sparse on implementation details, but the conceptual framework around when to remember versus when to forget is solid.