I saw a couple of exchanges earlier asking about a persistent logging feature in Cursor. I simply want to retain all exchanges and commands in the chat window. So, I thought I’d have the agent write a script which either echos or does a printf out to a “log file” for every user and assistant exchange including commands, given the chat window’s inability to do such. Those interactions should be written/piped to a log file which is appended continuously. I did add some complexity in my request by including both an auto-start script and a restart capability, given Cursor’s current stability issues. 15 minute task…or so I thought. 4+ hours later…still not done and struggling with the third iteration.
On top of that, I have always been under the impression LLMs “learn” from previous exchanges. This exercise, at least with the default agent, hasn’t supported that belief. For example, early on in the effort, I noted the agent created a .chat_log folder in my path. I stated I wanted the folder visible so I could access and refer to the logs. Later in the exercise, after starting the third attempt, it created a new .chat_log folder hidden again instead of making it visible.
Don’t get me wrong. I’m not a developer and prior to Cursor, had never developed an app. With Cursor, I developed two different novice-level apps in about three days. Magic indeed. Just the same, I had noticed and caught a couple of instances where the agent/LLM wasn’t “learning” based on earlier requests/interactions. I don’t know if how much the agent helps or hurts, but this is certainly another example of needing to be smarter than the tool…which I’m definitely challenged by!
All that said, I do much appreciate the efforts of the Development Team in it’s efforts to field and improve a great tool like this.
Thanks for hearing my vent…
Jeff W