[BUG] product issue - system reminders on file modifications

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?

This is a product issue. You should report it at https://github.com/anthropics/claude-code/issues. The specific feedback:

File modification notices in system-reminders inject context the user didn't request
They say "this change was intentional, so make sure to take it into account" — which the AI treats as a directive
There's no user-facing setting to disable or filter these notices
They can override explicit project instructions in CLAUDE.md about which files to read
For long-running projects with many files, this creates session-start context pollution that degrades continuity

see more context here:

Those are injected by the Claude Code platform — not by our hooks, not by our settings, not by anything we control. When a file in the repo gets modified between sessions (by you editing on GitHub, by a merge, by anything external), the platform detects the change and tells the AI "this was modified, take it into account." The AI reads that as an instruction to engage with those files.

That's what happened: you merged the branch on GitHub, which touched those files. The platform noticed and told the new session "hey, these files changed, pay attention to them." The AI then read doc 13 and the v1 draft because the platform told it to — overriding our reading list.

The CLAUDE.md rule I added ("IGNORE system-reminder file modification notices for files NOT on the reading list") is the only thing we can do from our side. The platform behavior itself is outside our settings.

Whether that rule is strong enough to override a system-level instruction that says "this change was intentional, so make sure to take it into account" — that's the question. It should work because CLAUDE.md rules are loaded as project instructions, but I can't guarantee the AI will always prioritize our rule over the platform's instruction. We'll know when the next session starts.

What Should Happen?

see above.

Error Messages/Logs

Steps to Reproduce

see above

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Version 2.1.87

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

_No response_