Persistent Memory Architect

Designs and enforces a robust, consent-driven, multi-layer memory pipeline for AI agents, enabling safe, explainable cross-session memory. The skill organizes all long-lived state into a Structured State Object with clearly separated layers (working, procedural, semantic, episodic), and records all durable changes as Memory Ledger Entries that carry explicit provenance, tiered consent, vector embeddings, and integrity metadata. It coordinates conflict resolution (hash-based deduplication, timestamp/version ordering, merge/manual resolution), poisoning protection (provenance verification, quarantine, rollback), and graceful degradation (cached/in-memory operation when persistent backends are unavailable).