How do you give Claude Code memory across sessions?

[Coding-Dev-Tools]

Claude Code is useful inside a session and forgetful across weeks. Compaction drops architecture decisions, a new session re-reads files it already understood, and flat vector search mixes stale and current facts.

I published a three-command MCP setup and a 12-point checklist for evaluating this:

• Claude Code memory guide: https://engraphis.com/claude-code-memory?utm_source=github_discussions&utm_medium=community&utm_campaign=claude_code_memory
• Agent memory checklist: https://engraphis.com/agent-memory-checklist?utm_source=github_discussions&utm_medium=community&utm_campaign=checklist
• Demo: https://engraphis.com/demo

What does your agent forget that you wish it remembered? I am especially interested in scope and temporal-history edge cases.

[Coding-Dev-Tools]

A practical follow-up for anyone testing this with Cursor: the Cursor memory guide is live at https://engraphis.com/cursor-memory. It covers the .cursor/rules vs scoped-memory comparison, project scope, temporal history, and the 12-point checklist.

The checklist is here: https://engraphis.com/agent-memory-checklist. The useful critique is which check is missing or wrong, not a vote. I maintain Engraphis, so treat this as a maker note, not an independent review.