I work primarily in multi-root workspaces as a cross-functional contributor. I’ve noticed that initializing a chat in a workspace of 3+ repos can often cause the agent to assume you are referring to one repo over another. For example:
Task: ”Apply the style guide on all articles touched on in this branch.”
Expectation: Cursor sees that i’m actively on repo x using branch y and chooses that as the starting point. We already know which repo + branch we’ve switched to in the bottom left of the IDE.
Reality: The agent often picks one of the other repos and executes the task, so I have to restart.
This is a relatively minor problem, but it does waste time and it feels preventable.