What problem does it solve? AI agents with flat-file or stateless memory lose context between sessions, cannot forget gracefully, and have no auditable record of what they learned. This Skill replaces basic memory with a cognitive architecture featuring episodic, semantic, procedural, and core memory stores. ## Core Features & Use Cases - Four Memory Stores: Episodic daily logs, a semantic knowledge graph with entities and relations, procedural workflow files, and a vault for pinned memories that never decay. - Decay-Based Forgetting: Relevance scores decay over time using a configurable half-life, reinforced by access counts, so stale memories fade instead of cluttering retrieval. - Reflection Engine: Sleep-time consolidation cycles that extract facts, rewrite core memory, and produce internal-monologue reflections with self-awareness tagging and a token reward system. - Multi-Agent Access & Audit Trail: Shared-read, gated-write model where sub-agents propose memories, plus git commits and audit.log tracking for every mutation. - Use Case: Set up a long-running personal AI assistant that remembers your preferences, consolidates daily conversations into a knowledge graph, and evolves its self-image over weeks of use. ## Quick Start Run the init_memory.sh script against my workspace directory, then enable memorySearch in my agent config and append the memory block template to AGENTS.md.
