I have observed some of the complaints on the forum and I believe some might be due to not the best prompts from the user. Wondering if there could be a feature that grades the quality of your prompt before you send it?
Also, I find it often very helpful to ask the agent a question, or describe a bug or feature, and instruct the agent to not write or change any code, but to just create a document that contains an analysis of the issue / dev plan. I almost feel like that should be a built in option, like a toggle in the prompt terminal. It help avoid the “bug” of “I asked cursor to do a simple thing and it made hundreds of changes”, because it gives you a chance to review the plan and give feedback. The plan then also provides great context for the model when do ask it to complete steps in the plan.