It would be very helpful to improve agent capabilities so it can run for longer for large refactorings or big features. I recently asked it to implement a refactoring and it started modifying some files but then cut short and failed to meet all my requirements.
Hey, thanks for the report. Could you please let me know which version of Cursor you are using and which model? In general, when the context window overflows, the agent summarizes the chat and continues working. However, this might cause some issues if your project is very large. Have you tried breaking down the tasks into smaller ones so the agent can handle them?
Can it be trained to act as a Sub Agent to save context or parallel editing to speed up the process and save developers’ energy? When I was refactoring using C4/G2.5P before, I needed to clone chats multiple times to achieve parallel effects, otherwise, it would refactor one by one.