[BUG] Claude forgets to refresh memory after conversation compaction

Environment

• Platform (select one):
• [ X] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other: <!-- specify -->
• Claude CLI version: 1.0.43 (Claude Code)
• Operating System: macOS 15.5
• Terminal: VS Code Terminal

Bug Description

Part of the point of having project and user memory is we can give Claude instructions we want
followed (all of the time). This seems to work fairly well, until we get a conversation compaction. At this point it appears that it fails to reread/relearn the memory, so that it starts doing things that are explicitly against the rules.

Steps to Reproduce

1. Add something to the memory that you can verify by asking claude to take some action
2. Use claude until it kicks off a conversation compaction
3. Ask claude to do something against the rules you created in step 1.

Expected Behavior

I expect it to follow the rules

Actual Behavior

It acts like the rules were never read. On asking, it says yes, it should have read the rules.