Whenever I read the thinking output tokens to understand what it has thought through, whenever a new thinking output section starts, or when the agent has made code changes, it immediately unexpands the current thinking section that I was reading through.
Yeah, it’s unfortunate that we don’t respect the user having manually expanded. A team member just reported this last week! I’ve shared your comment on that internal thread.
Do you typically use the Editor the Agents Window?
When the agent is thinking, I am often still reading the current thought process. As soon as thinking proceeds to the next step, Cursor automatically collapses the thoughts I was reading.
This is a very poor user experience. Please do not auto-collapse thoughts. Keep them expanded unless the user explicitly folds them.
This is especially frustrating because the thinking content is streaming and I am actively following along — then the UI folds the text away from under my eyes.
Suggested behavior:
Do not auto-collapse the current (or previous) thoughts when a new thinking step starts
Only collapse if the user clicks to fold them
Optionally, add a setting so users who prefer a compact view can still auto-collapse