Agent Memory En · fafofafo Build "Agent Memory" — a universal infrastructure giving any AI agent long-term memory, automatic learning, and self-evolution: cross-session memory, conversation auto-sedimentation (digest), hybrid semantic retrieval (BM25 + vector + Rerank), memory graph, pruning, and encryption. Use this skill when the user asks to implement/reproduce/build a memory system, or needs long-term memory, conversation sedimentation, semantic retrieval, knowledge accumulation, or self-evolution capabilities. Contains the complete implementation spec, key design decisions, and acceptance criteria — a new harness instance can reproduce it step by step from this document alone, no further explanation needed.
