Agent 长期记忆治理系统

分享海报

@user_638e256a/memory-governance

Agent long-term memory governance system for the agent-cognitive-brain L2 (short-term task memory) and L3 (long-term personality memory) layers. Provides TF-IDF semantic indexing with zero-dependency pure Python implementation, memory deduplication and merging (similarity above 0.86 threshold), hybrid recall router (keyword plus semantic with decay weighting), anaphora resolution for context backfill, memory retention policies (heat decay, TTL soft expiration, per-topic quota eviction), and health monitoring (fragmentation, redundancy, cold-ratio scoring). This skill should be used when building agent memory systems, adding semantic search, implementing memory deduplication, setting up context auto-backfill, or diagnosing memory health. Trigger keywords include: 记忆治理, 长期记忆, 语义检索, 记忆去重, 上下文回填, 指代消解, 记忆召回, Agent memory, TF-IDF.