Why is it getting worse day by day?**
It suddenly flags code that was already completed and confirmed to be bug-free days ago.(But code doesn’t have any errors)
While I’m in the middle of debugging a YAML task schedule, it randomly brings up unrelated past issues.
Sometimes, it doesn’t understand the question I asked just one message earlier — as if it’s receiving a brand-new request.
Fake Reporting.
Backed issues ?
Are you restarting chats frequently? When context windows get to big it causes issues like the ones you mentioned.
You mean new chat or restarting cursor app ? actually Im trying to use chat with sole issues to lessen problems . Mostly those issues are happened during refactoring.
Set the rule → Force it to read the necessary parts when running.
I’m not sure why, but it currently reads only certain parts of the code, which leads to many issues like fake code or inaccurate reports.
You should restarting the index base, and make sure there’s only one project in the workspace.
Yes right ! Fake report , errors codes are mostly caused by the assuming by something, and context seems manipulated easily ? Im not sure it can be connected to language options, because when I tried to use it various languages, the way operation is different.
Did you set rule in Cursor ?
Most of the problem lies in the understanding between you and the model, I think it’s best to have rules and try to give it a problem (Research + Reading + Plan)
I usually use 2 chat tabs, 1 chat tab to track progress. When AI1 finishes, I ask AI2 to check the code and evaluate it, then send it back to AI1 (currently I haven’t seen A2A integrated yet)
Have you tried writing down every requirement and desired feature in a file? I find forcing cursor to write everything down sometimes gets better behavior out of it. Ask for detailed reports and multiple solutions. Also ask it to find a solution instead of create a solution.
Actually it’s getting worse day by day. I don’t know their algorithm but it mixed up with information which is used in old chat. even there is no information on code and ref files. and sometime it end the editing after few trying and make fake reports. In I point out that fake reports, it start to do full analysis again ? Wow Token collector -_-;