[BUG] Desktop Remote Control: idle sweep stops a session seconds after it resumes, killing the turn mid-work (Windows)

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?

When a Remote Control session in the Claude desktop app has been idle for more than an hour, the idle sweep stops it, which is expected. When a new message then wakes it, the app resumes the session, but the idle timer is not reset. The next sweep, 30 to 60 seconds later, stops the session again while it is in the middle of a turn. The thread shows the worker as disconnected and the turn's work is lost. This repeats every time the session is woken.

The PC stays awake throughout on wired Ethernet. The Windows event log shows no sleep, crash or network events at those times, and there are no crash dumps.

What Should Happen?

Resuming a session, and any turn activity after that, should reset the idle timer. A session with a turn in progress should never be swept.

Error Messages/Logs

From the desktop app's remote-control.log (local time, session id redacted). In each case the session had just been resumed for a new message and was actively working. The logged idle time (86 min, 88 min, 16 h) is counted from before the resume.


16:56:25 attached (attached) session=cse_<redacted>
16:57:04 idle sweep: session=cse_<redacted> idle for 5168277ms
16:57:04 stopping session (idle)

16:58:16 attached (attached) session=cse_<redacted>
16:59:04 idle sweep: session=cse_<redacted> idle for 5288293ms
16:59:04 stopping session (idle)

09:36:10 attached (attached) session=cse_<redacted>
09:36:44 idle sweep: session=cse_<redacted> idle for 58483575ms
09:36:44 stopping session (idle)

Steps to Reproduce

1. In the Claude desktop app on Windows, run a Remote Control session driven from claude.ai.
2. Leave the session idle for more than 60 minutes until the idle sweep stops it ("stopping session (idle)" in remote-control.log).
3. Send the session a new message from claude.ai. The app resumes it ("admitted (resume, nudge poll)", then "attached").
4. While it works on that message, wait for the next idle sweep, about a minute later.
5. The sweep reports the pre-resume idle time and stops the session mid-turn. Sending another message repeats steps 3 to 5.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.284 (bundled with Claude desktop app 2.16120.0.0, MSIX). npm latest is 2.1.286, which the desktop app has not picked up yet.

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

This is different from issue #87777 (the 900s teardown on a session's first turn). Here the session has already been swept once for being idle, and every later resume gets swept again within a minute because the idle clock never resets.