Cursor with Claude has memory of a goldfish

Describe the Bug

Even when I have a prewritten prompt to remind it what not to do, it just does it again. The main mistakes it keeps making is it will edit the wrong css or js files even AFTER I’ve told it what file we are working on. Hilariously, it is now cussing back at its own retardation. My stress level just goes up with it.

Steps to Reproduce

Make it remember better because it forgets easily even after starting a new chat!

Screenshots / Screen Recordings

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.2.4 (user setup)
VSCode Version: 1.99.3
Commit: a8e95743c5268be73767c46944a71f4465d05c90
Date: 2025-07-10T17:09:01.383Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.26100

Does this stop you from using Cursor

Yes - Cursor is unusable

See, did it again. So unusable now.

I typically just reference previous chat with a screenshot so it just continues but “fresh” or reset. If you’re a pro member/on the new cursor beta etc. Have you verified the memories are being made with the proper memory tool and taking into consideration “which” LLM is writing/creating them? The one that makes them will only pertain to that one unless it can utilize its own “contextual awareness” and understand the project structure itself.

Of course if you you AUTO (CHATGPT) you will get only garbage code and the “OH MY GOD YOU WERE RIGHT”! with a bunch of stupid emojis…

Try to “train” them.

1 Like

I noticed if claude is in the flow and especially shooting for quick fixes, the memories totally get ignored! I will add more reference files not in the context window with every prompt… hope that helps.

1 Like

Yeah I understand that. The thing is though, it’s not even (just)Cursor. This has always been my experience that when programming complex backend stuff, it’s brilliant, but the second you throw custom CSS tasks at it, AI can’t see the way humans can just yet (even with browser MCP). 100% agree on never using “AUTO” because it’s a waste, but at this point, I’ve burned through $100 on Claude “4.0” just to deal with basic frontend customization and most of its mistakes I literally just gave up and fixed myself by finding the class/ID names.

1 Like

True. I found that it’s especially counter intuitive to think the longer the chat is the more memory the AI can remember - I mean it WOULD BE true if Cursor or whatever model company allocates that much for consumer level, but in practice, I find it best to just start a new chat with every small task, just so it doesn’t run out of memory that quickly.

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.