Cross-Harness Unified Memory: Build a Shared Neo4j Memory Layer Across Claude Code, Codex, and Cursor Using Standardized Hooks

Towards Data Science·medium signal

A Towards Data Science deep-dive demonstrates building a single persistent memory layer shared across Claude Code, Codex, and Cursor using hooks — shell commands that fire automatically on lifecycle events (SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop). Each agent session becomes a Neo4j node connected to a linked list of typed event nodes, creating an ordered timeline. The key insight: hooks are remarkably standardized across providers.