[BUG] CLAUDE.md/memory file edits show "Writing memory" without diff - cannot see what changed

Additional Corroboration from Twitter/X

Did some searching and found others noticing similar changes today (Feb 4, 2026):

"Writing memory" feature noticed:

• @0xChemistt reported Claude Code now writes to memory files - appears to be a stealth rollout without official announcement

Diff rendering issues:

• @jeetizee (Feb 3) reported broken diff rendering with "ghost artifacts"
• @SlavaVishnyakov observed drifts in diff handling around late Jan, noting Claude "became less transparent"

General behavior changes:

• Multiple users reporting "Claude Code feels different today" without corresponding version updates
• Community consensus seems to be these are server-side feature flag changes rather than client updates

This suggests the "Writing memory" display change is part of a broader server-side feature flag rollout affecting diff/memory handling. @ClaudeCodeLog tracks feature flags if that's helpful for investigation.

I also find this to be a huge regression. There is no way to tell what is being written to or read from CLAUDE.md, even hitting Ctrl+O and Ctrl+E doesn't show any more information. You have to manually go look at the file to see what changed.

Even more frustrating is the model seems to assume that the user can see it, so even when explicitly told to say what it just wrote to CLAUDE.md, it does things like this:

⏺ You're right, I keep editing the file without you being able to see the result. Here's the current state of the section:

⏺ Recalled 1 memory (ctrl+o to expand)

⏺ That's it — lines 73-79. Three bullet points under the heading "UI Component Philosophy." Does this look right to you, or do you want to adjust the wording?

So it tells you to hit ctrl+o but then ctrl+o just shows:

⏺ Read(CLAUDE.md)
  ⎿  Read 15 lines

Which is equally useless and frustrating.

There is also no distinction in the UI about whether the model is writing to CLAUDE.md vs MEMORY.md

This is a similar pattern to https://github.com/anthropics/claude-code/issues/21151 and feels actively hostile to any sort of oversight, introspection, or human governance.

@TweedBeetle Are you sure it was writing to CLAUDE.md?

I noticed Claude just started doing this, looked into and it seems it is writing to ~/.claude/projects/{project}/memory/. This seems to be a new memory feature (or the one which claude.ai has had for a while has now been enabled in Claude Code).

For now there appears to be no way to disable this feature, though you can block edits with permission rules.

More info here: https://gist.github.com/milobird/a1f256829ac75553c1a3bd64a5839eb4

@milobird from what I can tell writing to ~/.claude/projects/{project}/memory/MEMORY.md and any CLAUDE.md both just show the same opaque "writing 1 new memory" line. It also seems the project's MEMORY.md is inject into every conversation in that project, which seems like a questionable design choice.

@casepot Exactly.

I'm pretty sure it's a bug that this has been enabled in production, it's probably a feature that's still under development.

See also: https://github.com/anthropics/claude-code/issues/23341

Hi everyone, thanks for the feedback! I'm working on updates to the UI based on this thread as well as getting our docs for auto-memory published. If you want to opt-out, you can disable auto-memory by setting the env var CLAUDE_CODE_DISABLE_AUTO_MEMORY=1

@rboyce-ant The issue is the hidden diffs for CLAUDE.md updates, not auto-memory. I'd like to keep auto-memory enabled.

When I ask Claude to add a specific line to CLAUDE.md, I only see this now (no diffs). Which means I have to open the file and review every change manually. I'm using verbose mode too.

<img width="1315" height="587" alt="Image" src="https://github.com/user-attachments/assets/65b640e5-f8f4-48a5-9de2-3ccf67c71620" />

Hi everyone, thanks for the feedback! I'm working on updates to the UI based on this thread as well as getting our docs for auto-memory published. If you want to opt-out, you can disable auto-memory by setting the env var CLAUDE_CODE_DISABLE_AUTO_MEMORY=1

This has nothing to do with auto memory

The problem is WE CANNOT SEE what it has changed

Why do you treat your developers like retarded dumbasses? This is so annoying

After I tell claude to edit it's memory, now I need to tell it "what have you changed" just to see what it has changed, WASTING more tokens in the process

AT LEAST give us an option to disable this dumbass stuff you're adding for no reason that improves NO ones productictivity

Additional corroboration from X/Twitter

@novel_engineer (Feb 5):

"The memory implementation nerfs transparency in actually editing claude md files."

This directly echoes the core issue: the auto-memory UI treatment has been applied to CLAUDE.md edits too, hiding diffs that were previously visible.

Note: the broader auto-memory complaints (hidden MEMORY.md, not git-tracked, etc.) are a separate concern tracked in #23341. This issue is specifically about CLAUDE.md edit diffs being suppressed.

Adding a related data point: I have CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 set (the documented env var to disable auto-memory/MEMORY.md). The env var works correctly — nothing is written to MEMORY.md.

However, the UI still shows "Recalled 1 memory, wrote 1 memory" during normal sessions when Claude reads or edits CLAUDE.md files. This is confusing because:

1. The user has explicitly opted out of the "memory" feature
2. The UI uses the same "memory" vocabulary for both auto-memory (MEMORY.md) and CLAUDE.md file operations
3. Seeing "wrote 1 memory" when you've disabled memory creates the impression the env var isn't working

The root issue is that two distinct features — auto-memory (MEMORY.md) and CLAUDE.md read/write — share the same UI treatment ("Recalled/Wrote memory"). The env var only disables one; the other continues using "memory" terminology with no way to opt out of the collapsed UI.

Agreed with the original report that CLAUDE.md edits should show normal inline diffs like any other file. The "memory" UI abstraction is especially problematic for users who version-control their CLAUDE.md files and need to review changes before committing.

This is fixed as of v2.1.30 – CLAUDE.md changes will still show diffs like other files.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.