[BUG] Abnormal Token Consumption and Hard Limit Failure Leading to Unexpected $10 Charge (Claude Code)

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?

1. Issue Summary

While using Claude Code, my usage rapidly hit 80% of my plan limit within just 5 minutes after a session refresh. Furthermore, the usage exceeded the 100% threshold without triggering any automatic cutoff or safety guardrails, resulting in an unexpected and unauthorized $10 overage charge.

1. Occurrence Details & Account Info

Date & Time: Wednesday, April 29, 2026, approx. 14:00 ~ 14:30 (KST)

Location: Seongnam-si, Gyeonggi-do, South Korea

Environment: Claude Code opus 4.7

1. Timeline & Steps to Reproduce

~14:00 KST: Received a user permission request while executing tasks in Claude Code.

Action: I approved the permission request, which triggered a session refresh.

~14:05 KST (5 mins after refresh): While Claude Code continued its tasks, the usage abnormally spiked, reaching 80% of my plan limit in an unreasonably short time.

Subsequent Issue: The process did not halt upon reaching 100%. The system bypassed the hard limit, continuing to consume tokens and resulting in an automatic additional charge of $10.

1. Action Requested

Please review the backend and session logs for the specified timeframe to investigate the root cause (e.g., agent infinite loop, context window handling error).

Issue a refund or grant account credits for the $10 overage charge caused by this system failure.

Verify and fix the hard-limit safeguard to prevent this from happening again.

What Should Happen?

Expected vs. Actual Behavior

Expected Behavior: * A new session should not consume 80% of a plan limit in 5 minutes under normal operations; there should be protections against infinite agent loops or token calculation errors.

The system must enforce a strict hard limit at 100% usage and halt operations to prevent unexpected billing, unless explicit prior consent is given.

Actual Behavior: * A severe token leak or runaway loop occurred.

The billing guardrails failed to activate, leading to a $10 overage.

Error Messages/Logs

Steps to Reproduce

Timeline & Steps to Reproduce

Step 1: During an ongoing session in Claude Code, I entered the following prompt:

"이슈는 폐기하자. train_v3 학습해보자. local pc가 비어있어." > (Translation: "Let's discard the issue. Let's run train_v3 training. The local pc is empty.")

Step 2: Following this prompt, Claude Code requested user permission to proceed with the requested task (likely to execute the training or explore the local environment).

Step 3: I approved this permission request at approximately 14:00 KST, which triggered a session refresh.

Step 4: Within just 5 minutes of this approval (by 14:05 KST), the token usage skyrocketed to 80% of my plan limit. The incident occurred exactly between the messages before and after the prompt mentioned in Step 1.

Step 5: The agent continued running the task without hitting a hard limit, eventually exceeding 100% usage and causing an unexpected $10 automatic charge.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

opus 4.7

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Xterm

Additional Information

_No response_