Auto-debug is an agent for fixing errors in Cursor's terminal. To use it, click the blue "Auto-debug" button that shows up after a terminal error. When clicked, the AI will look through your files and enter a chain of thinking to try to fix the problem.

To some degree whenever AI has trouble understanding an issue it's because I don't understand the issue. And so that's on me to figure things out more in order to explain more.

At this stage, I'm writing my codebases to be LLM-friendly right from the start and designing around that, because why not

My general rule thumb has been, "If I require a context window larger than GPT4 is capable of supplying—while still being sufficient—there's a good chance the code needs to be more modularized/simplified." This has allowed me to resolve most issues that I had when interfacing with GPT as I was first starting out, and it hasn't led me astray since!