The Problem
You: “Remember that auth bug we fixed?”
Claude: “I don’t have memory of previous conversations.”
You: “We spent 3 hours on it yesterday”
Claude: “I’d be happy to help debug from scratch!”
200K context window. Zero memory between sessions.
You’re paying for a goldfish with a PhD.
The Fix
You: “What did we decide about auth?”
Claude: “We chose JWT over sessions for your microservices.
The refresh token issue - here’s exactly what we fixed…”
One file. Claude remembers everything.
We all know the problem, right? Would be great to have a brain for cursor as well!
Here is what I do, in a general sense, to give it some kind of memory:
I always make plans and fully iterate on them until they look good to me. I then save them to the project (which in my case is a git repo, and I commit all of my plans to the .cursor/plans directory). The plans then serve as a decision record that can be provided directly to a fresh new chat/agent. One or more can be provided, just like any set of files.
If you are working out of a project that is a git repo, prompt the agent to figure out what code changes were done by checking specific git hashes, comparing to the HEAD, etc. After that, it now has context of what work was done.
If you find yourself needing to do this a lot, then another thing you can do is at the end of an important point (perhaps you need to start work in a fresh chat) have the chat you are about to abandon create a file with a summary of what has been done; then you can save that out too and share it with another chat.
For situations where any of these can be combined, then you can prime a chat fairly well.