Agent stuck on “Taking longer than expected…” until user sends “continue”

[tieclip0506]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

In Composer with any agent( including GPT-5.3), the UI shows “Taking longer than expected…” and the run appears to hang. The model stops mid-edit (e.g. code fades out on an incomplete line). It does not recover on its own. The only reliable way I’ve found to unblock it is to send a short follow-up such as “continue” in the input box. This has happened many times, not a one-off.

Steps to Reproduce

1\Open Composer in Agent mode with Composer 2 Fast.
2\Run a task that reads/edits files (e.g. JSP/JS as in the screenshot).
3\Wait until the status shows “Taking longer than expected…” below the streamed output.

Expected Behavior

The agent should either finish the step, show a clear error, or time out and offer a retry—without requiring a manual “continue” prompt.

Screenshots / Screen Recordings

Operating System

Windows 10/11

Version Information

Version: 3.0.9 (user setup)
VSCode Version: 1.105.1
Commit: 93e276db8a03af947eafb2d10241e2de17806c20
Date: 2026-04-03T02:06:46.446Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.19042

Does this stop you from using Cursor

No - Cursor works, but with this issue

[deanrie]

Hey, this is a known issue. The team is aware and actively looking into it. Here are a few workarounds that often help:

1. Disable HTTP/2. Add this to settings.json via Ctrl+Shift+P > Preferences: Open User Settings (JSON):

"cursor.general.disableHttp2": true

Then restart Cursor.

2. Reindex the project. Cursor Settings > Indexing & Docs > Codebase Indexing > Resync Index. In the megathread, this fully fixed it for some users.

3. Network diagnostics. Cursor Settings > Network > Run Diagnostics. Share a screenshot of the results here if the first two steps don’t help.

Also, if it freezes again, copy the Request ID right away (chat context menu in the top-right > Copy Request ID) and send it in that thread. That helps the backend team investigate.

Main thread for this issue: Taking longer than expected

Let me know how it goes.