[FEATURE] Warn when auto-memory starts empty while populated stores exist for other paths

Open 💬 0 comments Opened Jun 14, 2026 by BobTells

Preflight

  • [x] I have searched existing requests — the global memory tier (#62026, #39195) and configurable storage (#28276, #66990) asks already exist. This request is narrower and, as far as I can tell, unfiled: warn the user when a session starts with empty auto-memory while populated stores exist for nearby paths.
  • [x] Single feature request.

Problem Statement

Auto-memory is scoped per git repo (documented/intended). But when you launch Claude Code from a folder that isn't inside a git repo, each launch path silently becomes its own separate memory store. There's no global fallback and — the actual problem here — no signal that anything is wrong when a session starts with an empty store while other, populated stores exist on the same machine.

The failure mode is invisible: saved corrections simply stop applying. The model behaves as if it never learned them, with nothing indicating the notes live in a different folder's store. I lost ~4 days re-hitting already-corrected mistakes before diagnosing it by hand — there was no in-product breadcrumb pointing at memory scoping as the cause.

This bites people who don't keep all their work inside a single git repo (a common setup for folks early in their journey). They have no way to discover that fragmentation is what's happening.

Proposed Solution

When a session initializes auto-memory and finds the current store empty, but detects other non-empty memory stores under ~/.claude/projects/, emit a one-time, low-noise notice — e.g.:

Auto-memory for this folder is empty, but populated memory exists for 3 other paths. If you expected your notes here, see autoMemoryDirectory to point them at one store.

The bar is just "don't fail silently." It doesn't need to merge or load anything — only surface that the situation exists and point at the existing fix (autoMemoryDirectory).

Alternatives Considered

  • autoMemoryDirectory (#28276, #66990) and a global tier (#62026) both fix fragmentation once you know about it — but neither tells you it's happening. This request is purely the missing detection/warning.
  • git init at the parent so subfolders share one repo store — works, but you only think to do it after you've already lost days.

Priority

High — significant, hard-to-diagnose productivity loss.

Feature Category

Configuration and settings

Use Case Example

  1. Launch Claude Code from C:\Work (not a git repo). Save several corrections over a week.
  2. Later launch from C:\Work\projectA (also not a git repo) — a different store, which starts empty.
  3. Claude no longer applies any of the prior corrections; nothing indicates why.
  4. With this feature, step 2 would print: "memory here is empty, but populated stores exist elsewhere — see autoMemoryDirectory," and the user diagnoses it in seconds instead of days.

Additional Context

  • Claude Code v2.1.141, Windows 11.
  • Related (already filed, not duplicated here): #62026 (global tier), #28276 / #66990 (autoMemoryDirectory), #34437 (worktree sharing).

View original on GitHub ↗