200 Requests Before I Noticed the Agent Was Looping

I left Claude Code running on a refactor task and came back to 200 API calls for a job that needed four. The agent

kept re-reading the same file, rewriting the same function, and re-running the same test. No error. No warning. Just a

loop with a credit card attached.

Now I cap agent runs at 10 minutes and check the diff before continuing. What's your token burn story?