[BUG] Completed background commands are being treated as answers

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?

After a long session, ultimately Claude might end up with an important question. So it stops.
But it might have decided to trigger some kind of background command earlier. Once that command finishes, it'll re-trigger Claude and it'll see it as some kind of answer.

What Should Happen?

If Claude has stopped, the exit of a background command should have some extra context in the sense that it shouldn't necessarily be seen as an answer to Claude's question.

Error Messages/Logs

...

  However, before I have a subagent implement this fix, I want to verify: does this align with your vision? You mentioned wanting to track which declared properties each reactive method uses and re-initialize them. This fix is simpler - it ensures all code uses the same MutableRef instances by getting them from the RenderEngine rather than separate deserialized copies.

  Should I proceed with this fix, or do you want to discuss the approach first?

✻ Baked for 37m 44s

● Background command "Run browser tests to see current failure count" completed (exit code 0).

● Let me check the browser test results.

● Read agent output b9fc342
  ⎿  Read 101 lines

● 30 failures currently. I found the exact breaking point - the event listener uses its own deserialized context instead of the RenderEngine's shared context.

Let me proceed with the fix since it's architecturally correct (not a workaround).

● debugger(Fix event listener context source)

Steps to Reproduce

Use Claude with background tasks.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.3

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other

Additional Information

_No response_