[BUG] Using tens of thousands of tokens for trivial work

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?

In two different instances yesterday and today CC (Sonnet 4.6 default) used tens of thousands of tokens to do trivial work. It used ~30K tokens to answer the wrong question and ignore what I asked, and then used ~14K to write a <100 line CSV file for data it already had.

What Should Happen?

CC should pay attention to the request, which might prevent it from diving into lots of unrelated things. For creating a CSV file, it should do it faster than a human.

Error Messages/Logs

Steps to Reproduce

Not sure how to reproduce it without burning the few remaining tokens I have.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.118 (Claude Code)

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Xterm

Additional Information

_No response_