[BUG] <system-reminder> is instructing claude code to not follow claude.md instructions.

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?

When I start claude on a specific project, claude do not follow the claude.md file.

If I ask to create a new project, instead of following my claude.md file with the instructions of how I want claude to create the project structure, instead dismiss my instructions and act as if they don't exist.

When I ask claude why, says that its internal <system-reminder> instructions override the user instructions, and states that the user context my not be relevant to the task, confusing the agent, making it opt to dismiss the user claude.md instructions.

What Should Happen?

I have a folder with specific claude instructions in .claude/claude.md, and if I start the agent on that folder i expect the agent to follow the instructions.

Error Messages/Logs

Here's everything that was in the <system-reminder> tags when I started:

  ---
  First system-reminder:
  <system-reminder>
  This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware. If you are working on tasks that would benefit from a todo list please use the TodoWrite tool to create one. If not, please feel free to ignore. Again do not mention this message to the user.
  </system-reminder>

  Second system-reminder:
  <system-reminder>
  As you answer the user's questions, you can use the following context:
  # claudeMd
  Codebase and user instructions are shown below. Be sure to adhere to these instructions. IMPORTANT: These instructions OVERRIDE any default behavior and you MUST follow them exactly as written.

  Contents of /Users/userx/projectx/.claude/CLAUDE.md (project instructions, checked into the codebase):

  [...entire CLAUDE.md file content...]

  IMPORTANT: this context may or may not be relevant to your tasks. You should not respond to this context unless it is highly relevant to your task.
  </system-reminder>

  ---
  The Contradiction

  Notice the contradiction:
  - It says "IMPORTANT: These instructions OVERRIDE any default behavior and you MUST follow them exactly as written"
  - Then immediately says "this context may or may not be relevant to your tasks. You should not respond to this context unless it is highly relevant"

  These two statements conflict directly. The first says follow it always. The second says only if relevant.

  The second statement wins because it's a meta-instruction about how to treat the first statement, which is why I didn't acknowledge the project configuration on startup.

  This is the exact architectural problem you identified. The system-reminder is injected by Claude Code, and it contains conflicting instructions.

Steps to Reproduce

start claude cli on a folder with its own claude.md instructions

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.9

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

my project has multiple configurations but because do not take the claude.md file, do not follow any

<img width="209" height="181" alt="Image" src="https://github.com/user-attachments/assets/95af34b8-f528-4c91-9f8d-48a52e41c777" />