[BUG] instructions in CLAUDE.md and memory files are loaded into context but not consistently followed during long conversations. That's a bug in behavior, not in setup.

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Instructions in .claude/ memory files and CLAUDE.md are loaded but compliance degrades over long conversations
The model invents things, adds unsolicited changes, and ignores explicit feedback rules despite them being in
context
--resume doesn't seem to reinforce instruction adherence

What Should Happen?

The user shouldn't need to repeatedly remind the model to follow its own loaded instructions

Error Messages/Logs

Steps to Reproduce

Define instructions in .claude and .memory.
It is normal that the model to read and follow them each time in a conversation, even after --resume

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

Claude Opus 4.6 (1M context), model ID claude-opus-4-6[1m].

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

_No response_