Desktop (Windows): long sessions stall with spinner; /feedback cannot be submitted from them; app stuck on engine 2.1.229

Status Open
Reported on v2.1.229
Maintainer reply None cached
Activity 1 comment · opened Aug 17, 2026

Summary

On Windows Desktop app, long-running sessions stop making progress: the spinner keeps animating and the session appears to "read" indefinitely, but no work advances. Stopping the current action and re-sending a message temporarily resumes it, then it stalls again. Rebooting does not help.

Separately, /feedback cannot be submitted at all from such a session — it fails with "피드백을 보�낼 수 없습니다. 다시 시도하세요." (Failed to send feedback. Please try again.). Since the feedback payload includes the current session transcript, a large session appears to make the bug report itself unsubmittable. This is why I am filing here instead.

Environment

| Item | Value |
|---|---|
| Surface | Claude Desktop app (Windows) |
| Claude Code engine | 2.1.229 |
| OS | Windows 11 Pro 26200 |
| Node | v24.16.0 |
| Session size | Very large — multiple dynamic workflows (100+ subagents total), ~20 image reads (rendered PPTX slides), long tool-heavy history |

Steps to reproduce

  1. Run a long Desktop session with heavy tool use (workflows, many image reads, large file operations).
  2. Continue working in the same session for an extended period.
  3. Send a new message.

Expected

The session processes the message and streams a response.

Actual

  • Spinner animates, session appears busy, but no progress is made.
  • Interrupting the current action and re-sending resumes work briefly, then it stalls again.
  • Reboot does not resolve it; the behavior returns.
  • /feedback from this session fails to submit (see above).

Notes / possible relation to existing fixes

The v2.1.232 changelog contains an entry that matches this symptom closely:

Improved fullscreen streaming: long sessions stay responsive because the whole conversation is no longer re-normalized on every update

If that is the same root cause, two follow-ups may still be worth addressing:

  1. /feedback is unusable exactly when it is most needed. The report bundles the session transcript, so the larger and more broken the session, the less likely the report can be sent. Consider truncating/streaming the transcript, or failing over to a metadata-only report.
  1. The Desktop app is not picking up newer engine builds. On this machine %APPDATA%\Claude\claude-code\ contains only 2.1.229, while npm/GitHub already publish 2.1.233. No update prompt appears in the app, so the user is stranded on the affected build with no in-app path to the fix.
%APPDATA%\Claude\claude-code\
└── 2.1.229          <- only version present

npm  @anthropic-ai/claude-code  ->  2.1.233
GH   anthropics/claude-code     ->  v2.1.233 (2026-08-14)

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗