Model provider error lossing context/conversation but re-actived by sub task

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

GLM 5.2 always report unable to reach the model provider. In every task. can’t bear in a long task, then forget all the context and progress. if the model create some async task or sub model, then the async task or submodel trigger/active the session when they finished even I never use. That’s really bad while I’ve start new session/conversation to do the same task but the discard session actived by some async tasks and make my repo messive.

If model provide error, do not missing all the context/conversation and shutdown all sub task avoding re-active the main session or messing my repo

Steps to Reproduce

Just create new task and then happend, if not appear, just create new task. and there must be some sub task like async task or sub agent work.

then meet the model provider error.

then some how the sub task finished, the session reactived again but the main session forget all the context and conversation but just reactived by the sub task and the main session analysis the artifacts by the sub task and do next step without any context, randomly distribute instruction

Screenshots / Screen Recordings

clipboard_2026-07-01_16-03.png
clipboard_2026-07-01_16-03.png1520×228 31 KB

Operating System

MacOS

Version Information

Version: 3.9.16
VS Code Extension API: 1.105.1
Commit: 042b3c1a4c53f2c3808067f519fbfc67b72cad80
Date: 2026-06-27T06:41:01.941Z
Layout: glass
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Darwin arm64 25.5.0

Does this stop you from using Cursor

Yes - Cursor is unusable

5

BTW: the GLM 5.2 too many error about model provider error. I don’t know it’s cursor’s adaption problem or the model’s company problem/the model’s itself promblem. Critical bug; unusable without a fix.

6

Hey @Anudorannador,

Thanks for flagging this. On our backend, we’re seeing a higher-than-expected error rate for GLM 5.2, and the team is aware of the issue and looking into it.

9

Hi @Colin
GLM’s problme is not the point. Please focus on that stop re-activing session by sub-task finished but model report error and context and conversation lose either.