I think giving the option for it to be able to read the actual directory only in cursor app would be nice. It would be way more accurate if it could just read the current project folders and files then fulfill the request based off the tree and instructions given. There are a lot of times it randomly just creates a duplicate file of something because I didn’t manually give it the path each time (or simply just bypassing the path occasionally because it already has a lot to do). So the option to end it, an option for it to read the directory only, and an option for it to read and write to the directory of the current project folder would be a great feature to add. It would also cut down on all the copying and pasting going into the chat and composer by just telling it what files need attention.
This would be a significant feature, in my opinion, as well. Also, the ability to set/reset context refresh/resync when making hundreds of changes, as it does lose context (expected) after some time, and you have to start over. In using a repo (GitHub), I have found that it is also best to create a companion WIKI and then provide those URLs at the start of any session to provide the context. However, it would be much better if it “just did” this.
Thanks for the likes guys!!! I hope this becomes a thing. Right now I have to use tree /f /a > tree.txt to create a tree view in PowerShell of my current directory, once I have the tree view from tree.txt file, I drop that in the chat or composer in order for the ai to understand the whole project.
Your right, a set/reset context would be great thing because I have to drop this tree view occasionally to remind it. Not sure what the linux version of this would be but it does make it a lot easier when the ai understands the whole structure even if its just for a moment. You can tell in the way it outputs the code after that it understands a lot more with you drop it a tree view. I feel its a bit more accurate. So having the feature to read the directory itself would be such a great feature.
Aider has had this for a while and I honestly can’t believe cursor doesn’t yet, despite me requesting it several times. It’s stupidly obvious
OMG +10000000
So many times X is missing
to which there are 2 options
A. Create the new file
B. Correct the AI via a prompt - which then (if carrying out multiple steps will recreate the steps).
I shouldn’t have to correct the AI though (AND it shouldn’t rely on history alone to remember there was a file there).
A simple option would be to include the tree output of the folder into a readme.md file.
Then every time a new file is created or deleted update the file tree readme.md with the new structure.
Then add to the default prompt, always use the tree readme.md to look if the file exists first THEN add that to the codebase context.
Not sure why there is a hard limit on codebase size (to me it seems like a good pro/premium upsell feature to offer an increased codebase size) as the AI doesn’t know what needs to be in the context before it’s looked at the codebase based on the prompt (other than the explicitly declared files in the prompt)