I have been using Claude Code for a while now, and I have run into the following question recently: when Claude saves something to its auto-memory, should you commit that to git?

The short answer is: no - and you shouldn’t need to.

How Claude Code memory works

Claude Code has two distinct memory layers:

Auto-memory (~/.claude/projects/<repo>/memory/) is where Claude writes notes to itself as it works: things it learned about your project, conventions, corrections you gave it. This lives on your machine, outside the repository. It is never committed to git.

Project memory (CLAUDE.md / AGENTS.md) are instructions you write explicitly for Claude (and other AI tools). These live inside the repo, are versioned with git, and are loaded at the start of every session.

The rule of thumb

If a rule matters for the project, put it in AGENTS.md. If it’s a personal preference for how Claude behaves with you, auto-memory is fine.

The distinction is: would another developer or AI agent need this rule? If yes, it belongs in the repo.

A concrete example

Yesterday I asked Claude to save a rule to always generate a permalink front matter field when creating new posts together with me. Claude saved that to auto-memory.

But after questioning whether that was the right place, we concluded it’s a project rule — it affects anyone (or any AI) working on this repo. So we moved it to AGENTS.md and deleted the memory entry. Now it’s versioned, visible in code review, and available to every tool.

A few other things to pay attention to

  • Keep AGENTS.md under 200 lines — the official docs warn that longer files consume more context and instruction adherence measurably drops.
  • You can disable auto-memory entirely with CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 if you prefer to manage everything manually.
  • If your repo uses AGENTS.md, you can have CLAUDE.md import it so both tools read the same instructions without duplication.

References