Harness Memory Architecture

Claude Code's memory architecture works like this:

1. The user can create memories at any time with a slash command.

2. Memories go into a folder and then the memory index file is updated with a link to each memory.

3. The memory index file is injected into the system prompt.

4. On each turn extract memories from the conversation into memory files. GOTO #2

5. Memories are consolidated in the background once a day.