Most agent systems fail at memory because they treat it like document retrieval. This skill walks you through building proper memory architectures, from working memory in the context window all the way to temporal knowledge graphs that track how facts change over time. The benchmark data is genuinely useful: Zep's temporal graph hits 94.8% accuracy versus 60-70% for basic vector RAG, with concrete latency numbers included. You get the full spectrum of implementation patterns, from session-scoped storage to entity tracking to validity periods. If you're building agents that need to remember things across sessions or reason over relationships between entities, this covers the architectural decisions you actually need to make.
npx skills add https://github.com/shipshitdev/library --skill memory-systems