A model may update a file multiple time. Gemini (reported above) has sometimes an tool call issue which is model related. Forwarded to Cursor Team.
i wouldnt stack many changes before checking every now and then jaaj
i use claude 80% so it happens with claude a lot.
Yeah I am reviewing the code and stopping immediately if I see its doing wrong
1 Like
OP didnt stack changes 
Those are To-dos. Can be interrupted with a new corrective prompt push. oops, thats a feature too (Queue)
Well, finally updated.
I’m finding agent interaction exceptionally slow, sonnet 4 takes long to respond and sometimes thinking for over 30s for simple requests (a test request).
I don’t know if it’s really related, I didn’t check before, but in 1.2, the close chat icon is on the chat icon, so if I change tab, to come back I close the chat 1 time out of 2.
@Krosnoz do you have some other toolbars active that overlap?
here is default v1.2
@T1000 I used to have the labels, but now they’re grayed out.
All the other icons don’t have this problem because they can’t be closed with the hover.
EDIT: if you have the Activity Bar on top, the label won’t show, that’s the problem
EDIT 2: if you have this problem, you need to reclick here (Default) when update
@erophames The app and Agent should be faster than previous versions. What does Network Diagnostic say? Do you have a non-privacy Request ID I could forward?
Hello. I have feedback for to-do lists.
Problem:
When working on complex, multi-step coding tasks with the AI assistant, detailed progress within individual tasks gets lost when conversations exceed the context window. This forces users to recap design decisions and implementation progress, reducing efficiency.
Specific Example
- Start “implement auth middleware” task
- Have detailed discussion about JWT vs sessions (15+ messages)
- Make key security decisions and implementation choices
- Context window truncates or new session starts
- AI loses all nuanced progress, restarts from scratch
Proposed solution - enhance the todo/task system to include:
- Detailed progress notes per task
- Design decisions made during implementation
- Code snippets and approaches tried
- Blockers and their resolutions
- Requirements clarified during development
This would allow AI to resume complex tasks intelligently rather than starting over
What would be even better, is the ability to spawn sub-agents for each of the todo tasks! That would allow for better context management and top-agent supervision.
1 Like
How can I Queue?
Thanks
I see, items your screenshot shows: Problems, Terminal … are usually in Panel (bottom). When i right click i get only following options.
You can send more than one cue at a time and then it’s queued up
2 Likes
Can the team please clarify how to customize the marketplace used by Cursor? The instructions in VSCodium no longer work in v1.2
Please note that I’m not asking for instructions on using a particular marketplace (such as the VSCode marketplace), I’m asking for how the marketplace in use can be customized in Cursor.
its on by default just keep asking for tasks while its working.
Can you elaborate on customizing marketplace? What would that involve?
Where do I find the network diagnostic? I’m mostly a Visual Studio user.
1 Like
@erophames Cursor Settings (top right gear icon) > Network > Network Diagnostics