Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
tl;dr There’s a bug in agent command output parsing that reports itself as a connection error, thereafter breaking the chat session and potentially re-breaking future sessions.
There is a specific set of issues here that needs to be addressed rather than worked around by the end user, because it can render Cursor unusable and/or is disruptive. This is what I’ve observed (OP can confirm):
Some agent-run commands seem to take inordinately long and then report connection timeouts in the UI. It may be model-dependent, but I use Auto as much as possible. For me, it happens most when running tests with failures for React / Jest, where the test output contains a bunch of DOM elements. It could be a coincidence or something to do with parsing. In the dev tools log, I see an auto-bug report generated for request ID 35354a4f-1173-43d8-853c-efde74c44176. Hopefully someone at Cursor can pull that up and look.
This is the actual stack trace. It is not really a connection timeout, despite the popup:
workbench.desktop.main.js:2750 \[composer\] Error in AI response: undefined ConnectError: \[internal\] Serialization error in aiserver.v1.StreamUnifiedChatRequestWithTools
at workbench.desktop.main.js:4989:396431Caused by: Error: invalid int 32: 4294967295
at u6n (workbench.desktop.main.js:441:32634)
at N2a.int32 (workbench.desktop.main.js:442:12763)
at uMt (workbench.desktop.main.js:442:19797)
at B2a (workbench.desktop.main.js:442:18900)
at Object.writeMessage (workbench.desktop.main.js:442:16664)
at ssi.toBinary (workbench.desktop.main.js:441:34677)
at U2a (workbench.desktop.main.js:442:19718)
at B2a (workbench.desktop.main.js:442:18998)
at Object.writeMessage (workbench.desktop.main.js:442:16664)
at Iti.toBinary (workbench.desktop.main.js:441:34677)
at U2a (workbench.desktop.main.js:442:19718)
at B2a (workbench.desktop.main.js:442:18998)
at Object.writeMessage (workbench.desktop.main.js:442:16664)
at sai.toBinary (workbench.desktop.main.js:441:34677)
at workbench.desktop.main.js:4989:396402
Thereafter, if I click “Try Again” it will usually try again, possibly incurring additional charges, possibly undoing a lot of work, and almost always getting to the point of needing to run the tests and repeating the behavior, leaving me stuck.
If I click “Resume” It will sit literally forever in “Planning next moves.” Looking at the console at this point, it appears as though it fed the same problematic context into the model, got the same stack trace, but instead of translating the stack trace into a popup connection error, it swallowed the error. At this point, the chat is no longer usable. Forking the chat is no longer possible.
Closing and reopening Cursor temporarily fixes the problem (I believe it rolls back the context), but again, if it steps on whatever landmine was causing the issue, I’m in the same boat.
Steps to Reproduce
I do not have repeatable steps. It seems possibly peculiar to one model (I use Auto) and triggered by particular command output. I will gladly provide request ids or other diagnostics
Expected Behavior
Ideally all command output should be sanitized and accepted by all models. As a fallback, failure in output parsing should fallback gracefully and perhaps instruct the model as to what happened so it can avoid that command.
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Version: 2.0.77 (system setup)
VSCode Version: 1.99.3
Commit: ba90f2f88e4911312761abab9492c42442117cf0
Date: 2025-11-13T23:10:43.113Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.22631
For AI issues: which model did you use?
Auto
For AI issues: add Request ID with privacy disabled
35354a4f-1173-43d8-853c-efde74c44176
Additional Information
This has been happening through several versions. I believe others are seeing it but reporting it as connection issues.
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor