First time building an llm agent, student level. Pretty sure somebody have made similar things and perform much better. Just want to know what to improve.

Btw, if you try it, use some small repo. It seems can take really long time dealing with large repos and easy to fail. Still working on that.