[Bug] Claude ignores persistent user rules for build process restrictions

Your analysis is spot-on. The root problem is that memory rules are soft instructions, not hard constraints. The model can under-weight or overlook them during tool planning.

The fix already exists — use PreToolUse hooks

Claude Code's hook system provides exactly the enforcement boundary you're describing. A PreToolUse hook for the Bash tool can:

1. Intercept every bash command before execution
2. Match it against blocked patterns (cmake, make, npm build, etc.)
3. Exit code 2 → block the tool call entirely and explain why

This turns a soft memory rule into a hard execution guardrail — enforced at the tool layer, not dependent on model reasoning.

I've created a ready-to-use hook example that does this:

• Blocks build-system commands (cmake, make, ninja, gradle, npm run build, etc.)
• Allows non-build commands to pass through normally
• Shows a clear message explaining why the command was blocked

You can use it by adding to your hooks config:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "python3 /path/to/block_build_commands_hook.py"
          }
        ]
      }
    ]
  }
}

See the existing example at examples/hooks/bash_command_validator_example.py and the new block-build-commands example in PR #62264 for reference.

Got it working, thanks. Also added a bit to the output so claude not only says it's blocked from running the command but it outputs exactly what it attempted to run. Allows for an easy user copy/paste to run the command themselves.

The bit I added is in the 62264 conversation

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.