Claude Code Field Guide
14 / 35

Auto Memory

Claude Code notices your patterns and preferences and remembers them across sessions on its own — you can review and edit what it stored.

Updated since the video
Currency check: The concept is accurate; the mechanism is now far more formalized and inspectable than the video suggests.

CLAUDE.md is what you write down deliberately. Auto memory is different: Claude Code recognizes recurring patterns — you always work in a particular language, prefer a certain style, have specific project conventions — and saves that automatically, without you writing a file. It's not the same as conversation history; it's cross-session, generalized learning about how you work.

What's grown since the video: auto memory is no longer an invisible black box. It's organized into four kinds of notes — your role/preferences, corrections you've given, ongoing project context, and external references — stored as plain markdown files you can browse, edit or delete with /memory. Claude deliberately skips anything it can already derive from your code or that CLAUDE.md already states, so the two systems don't duplicate each other.

Corrections that actually stick

After correcting Claude Code twice for using npm instead of the project's pnpm, it saves that as a feedback memory — the third session gets it right without a third correction.

Auditing what Claude has learned about you

A privacy-conscious user periodically opens /memory to review exactly what's been saved, deleting anything project-specific that no longer applies after a role or team change.