[BUG] Claude Code ignores AGENTS.md rules and doesn't act as gatekeeper

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?

Problem:

Claude Code does not reliably read project rule files (AGENTS.md, CLAUDE.md)
Even when it claims to have read them, it ignores explicit rules like "Do not modify this file unless user explicitly requests"
Claude Code modifies backend files (controllers, API functions) without asking, even when rules forbid this
No gatekeeper behavior - it acts first, doesn't stop to ask

Comparison:
Codex (OpenAI) correctly:

Reads documentation files
Follows project rules
Stops and asks before making changes
Produces correct API/function code

Impact:

Broken backend code (400 errors)
User has to review every change manually
Trust is lost - user cannot rely on Claude Code for backend work

Expected behavior:

Mandatory reading of AGENTS.md at session start
Actually enforce "stop and ask" rules
Don't override user-defined restrictions

What Should Happen?

Problem:

Claude Code does not reliably read project rule files (AGENTS.md, CLAUDE.md)
Even when it claims to have read them, it ignores explicit rules like "Do not modify this file unless user explicitly requests"
Claude Code modifies backend files (controllers, API functions) without asking, even when rules forbid this
No gatekeeper behavior - it acts first, doesn't stop to ask

Comparison:
Codex (OpenAI) correctly:

Reads documentation files
Follows project rules
Stops and asks before making changes
Produces correct API/function code

Impact:

Broken backend code (400 errors)
User has to review every change manually
Trust is lost - user cannot rely on Claude Code for backend work

Expected behavior:

Mandatory reading of AGENTS.md at session start
Actually enforce "stop and ask" rules
Don't override user-defined restrictions

Error Messages/Logs

Steps to Reproduce

Problem:

Claude Code does not reliably read project rule files (AGENTS.md, CLAUDE.md)
Even when it claims to have read them, it ignores explicit rules like "Do not modify this file unless user explicitly requests"
Claude Code modifies backend files (controllers, API functions) without asking, even when rules forbid this
No gatekeeper behavior - it acts first, doesn't stop to ask

Comparison:
Codex (OpenAI) correctly:

Reads documentation files
Follows project rules
Stops and asks before making changes
Produces correct API/function code

Impact:

Broken backend code (400 errors)
User has to review every change manually
Trust is lost - user cannot rely on Claude Code for backend work

Expected behavior:

Mandatory reading of AGENTS.md at session start
Actually enforce "stop and ask" rules
Don't override user-defined restrictions

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.0.74 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_