### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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? Title: "Watch live run" link showed a stale/unclear traceback, couldn't tell if a background task was actually failing or not Body While running a backg

[BUG] "Watch live run" link showed a stale/unclear traceback, couldn't tell if a background task was actually failing or not

Status Open

Reported on v2.1.150

Maintainer reply ✓ Yes — bcherny

Activity 4 comments · opened Aug 6, 2026

💡 Likely answer: A maintainer (bcherny, collaborator) responded on this thread — see the highlighted reply below.

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?

Title: "Watch live run" link showed a stale/unclear traceback, couldn't tell if a background task was actually failing or not

Body

While running a background task (a Python eval script making live Anthropic API calls), I clicked the "watch live run" link to check on progress. It showed a Python traceback / crash. I asked Claude to explain, and it checked the actual output file for that specific background task and found no trace of that error — only normal output. It also couldn't fully explain why the "watch live run" view showed something the underlying task output file didn't contain.

I couldn't get a clear, trustworthy answer on whether the task I had running was actually failing or not, whether the link was showing current or stale/historical output, or how much I'd actually be charged for the API calls already made before I killed it. That's a real problem when the background task is spending real money per call — I need to be able to trust that what "watch live run" shows me reflects the actual current state of the task I'm looking at, not something ambiguous or from a different/earlier run.

Suggested fix: make it unambiguous which task a "watch live run" view is attached to (task ID, start time, running/finished status clearly labeled), and make sure live output shown to the user is guaranteed to match the actual current process's output — not a cached or unrelated view.

I need reimbursement on my wasted credits. its expensive as is, and I'm trying to find a job, and building things for propective employers' test challenge, and I can't afford to keep wasting credits thanks to Claude Code.

What Should Happen?

Fix this issue

AND reimburse me my credits.

Error Messages/Logs

Steps to Reproduce

Ask Claude Code to run full eval tests with adversarial tests ~190 runs, and when click watch live run.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

Claude Code version 2.1.150

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗

4 Comments

bcherny collaborator · 14 days ago

Thanks for the report. I tried to reproduce this on Claude Code 2.1.233 (macOS, Terminal): I asked Claude to run a Python script as a background task and then inspected the UI. Background tasks in the Claude Code CLI show a task ID, an output file path you can tail -f, and the /tasks list — there is no "Watch live run" link in the terminal UI, in this or any earlier version I could find, so I couldn't reproduce what you saw.

That link most likely came from something else — for example your eval script or an eval/observability platform it reports to, which would explain why its traceback didn't match the Claude Code task's output file.

To help us dig further, could you share:

Where exactly the "Watch live run" link appeared (a screenshot of the terminal would be ideal), and what URL it opened
The exact output of claude --version and claude doctor
Whether you were using the terminal CLI, the VS Code extension, or Claude Code on the web/desktop app

On the credits question: reimbursement isn't something we can handle through GitHub issues — please contact Anthropic support at https://support.claude.com so they can look at your account.

🤖 Generated with Claude Code

github-actions[bot] · 14 days ago

We weren't able to reproduce this. Could you provide steps to trigger the issue — what you ran, what happened, and what you expected? This issue will be closed automatically if there's no activity within 7 days.

pkotecha-eng · 14 days ago

Thanks for looking into this. Please forget about the reimbursement. I was just upset in the moment.

Client: Claude Code desktop app on macOS.
Version: claude --version → 2.1.150 (Claude Code). Noticed via claude doctor that this is a few versions behind both stable (2.1.224) and latest (2.1.233) — possible the behavior changed since. doctor also flagged a leftover npm-global install alongside the native one, so it's possible a different version was active when I originally saw this, though I can't confirm that.
The link: appeared after starting a background task, opened as a right-side panel within the same window showing live session updates — it purely in-app with no separate address
_ I'll try to rerun it now to see if it reproduces.