[BUG] Inconsistent documentation about "# to memorize" and CLAUDE.md persistence

Environment

• Platform (select one):
• [X] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other: <!-- specify -->
• Claude CLI version: 1.0.22 (Claude Code)
• Operating System: macOS 15.5
• Terminal: iTerm2

Bug Description

Claude Code provides inconsistent information about how # to memorize comments work, particularly regarding automatic updates to CLAUDE.md files.

claude showed me "Tip: Use # to memorize shortcut to quickly add to CLAUDE.md" after reporting on an update it made to CLAUDE.md

Yesterday, another claude session told me that "# to memorize" only causes flagging importance of a command inside the current session, but not across separate sessions, and that I needed to write something like "# to memorize - add to claude.me" to get claude to actually persist my instructions to CLAUDE.md.

So I asked claude today "which is it?" and was told:

⏺ You're absolutely right, and I apologize for the confusion! I was inconsistent and incorrect.

The truth is: # to memorize by itself does NOT persist across sessions. It only flags something as important within the current conversation.

To actually get instructions added to CLAUDE.md across sessions, you need to be explicit:

• "# to memorize - add to CLAUDE.md"
• "# remember this in CLAUDE.md"
• Or directly ask me to "add this to CLAUDE.md"

I should not have written "Tip: Use # to memorize shortcut to quickly add to CLAUDE.md" because that's misleading. The # to memorize flag alone doesn't automatically update CLAUDE.md files.

Thank you for catching this inconsistency - it's important that I give you accurate information about how these features actually work. I'll be more careful about this distinction going forward.

Steps to Reproduce

1. See the "tip" output when claude reports on an update to CLAUDE.md
2. Use # to memorize in a variety of situations. Notice that some do not persist across session (restarts of the claude command without the --continue option)
3. Ask claude itself about the inconsistency and see that its answer contradicts the tip

Expected Behavior

Ideally: "# to memorize" always causes an update to CLAUDE.md, as that appears to have been the intent
Less ideal: At least stop misleading people with a clearly false "top" that claude itself says does not work when asked.

Actual Behavior

claude only persists "# to memorize" instructions part of the time. This is truly confusing and frustrating.

Impact

• Creates user confusion about core functionality
• Leads to unreliable expectations about feature behavior
• Affects workflow planning and documentation strategies

Additional Context

This inconsistency was observed across multiple sessions where Claude gave contradictory explanations of the same feature.

LOVING THIS TOOL!
Fix this will only make claude even more awesome.
Thank you!