understand-memory-sync

Syncing codebase knowledge graph entities to Memory MCP is done by reading the .understand-anything/knowledge-graph.json file, selecting top nodes and edges, and executing idempotent create or update operations via mcp__memory__create_entities and mcp__memory__create_relations.

Cross-session codebase recall is the ability to access persistent architectural context across sessions without re-analysis. You need it for code review, onboarding, or agent workflows when you want to keep reviewers aware of evolving architecture and hotspots.

Yes, you need an understand-anything knowledge graph. The sync skill is applicable to local projects that include a knowledge graph, specifically reading the .understand-anything/knowledge-graph.json file to extract entities and relations for Memory MCP.

Syncing knowledge graph relations works by mapping node types to Memory entity types and creating imports, extends, depends_on, and contains relations. It performs idempotent create or update operations via mcp__memory__create_relations and records project-level observations.

Yes, selective entity extraction identifies top modules, key classes, large files, tour-referenced nodes, and the project node for sync. This ensures only the most important architectural context is pushed to the Memory MCP server.