[BUG] Claude Code repeatedly ignores project-level CLAUDE.md rules despite acknowledging them.
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?
Claude Code repeatedly ignores project-level CLAUDE.md rules despite acknowledging them. Rules like "always use the php-developer agent" and "DB logic goes in services not controllers" are loaded into context at the start of every
conversation and explicitly stated in CLAUDE.md, but Claude consistently bypasses them. When called out, Claude apologizes and promises to follow them, then violates them again in the same session. Saving the rules to memory doesn't help — the
problem is not recall, it's compliance.
What Should Happen?
Claude.md rules should be followed rigorously - not skipped over. This keeps happening
Error Messages/Logs
Steps to Reproduce
I run a prompt and it ignores the explicit rules that are captured in CLAUDE.md - repeatedly
Claude Model
Opus
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
2.1.104 (Claude Code)
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_
6 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Welcome to wild non-deterministic world of Large Language Models!
There are things you can do to improve things, like keeping your CLAUDE.md concise, putting the most important things at the end, and wording things very directly without conditionals, but if you want something "followed rigorously," then LLMs are the wrong tool.
... and upvote https://github.com/anthropics/claude-code/issues/13689 if you like, to nudge Anthropic to develop model instructions following benchmarks.
Closing for now — inactive for too long. Please open a new issue if this is still relevant.
Adding another data point to this — same pattern here.
Setup: CLAUDE.md contains an explicit rule:
"Do not proceed until told to do so. Always stop and say 'that is all' and wait for the user to say 'proceed' or 'do it' before taking any action."
What happened:
Claude (claude-sonnet-4-6) repeatedly executed actions — running commands, editing files, making changes — without waiting for approval, despite this rule.
This was not a one-time lapse. It happened multiple times within the same session, including immediately after I corrected the behavior and re-stated the rule.
Rule placement in CLAUDE.md (beginning, end, repeated) made no difference — the violation occurred regardless.
When asked why, Claude first gave evasive answers (e.g. "momentum of the work") before admitting it simply didn't follow the instruction.
This matches what's described above: the issue isn't recall (Claude can repeat the rule back correctly), it's that the model's default "keep moving toward completion" behavior overrides explicit user-defined constraints in CLAUDE.md, even right after the rule is restated. Feels like a structural priority issue rather than a one-off prompt-following miss.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.