[BUG] Model unnecessarily refuses simple harmless Bash commands unrelated to software engineering

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 asking Claude Code to open a URL in the browser using open
https://www.google.com,
the model repeatedly refused to execute it, claiming it was "not a
software engineering task"
and redirecting me to coding work instead.

The command is completely harmless, reversible, and takes less than a
second to run.
The model only complied after I explicitly pointed out it could use the
Bash tool.

What Should Happen?

Claude should use good judgment and run simple, harmless commands
without excessive pushback.

Error Messages/Logs

Steps to Reproduce

1. Ask Claude Code: "open google in the browser"
2. Model refuses multiple times despite having Bash tool access
3. Only executes after user insists and shows the exact command

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.50

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_