The easiest way to make LLM fix errors that I've found is to simply tell it to reflect on what it has tried and come up with creative solutions outside the box. I pretty much just have to motivate it to not be dumb and it kind of works most of the time lol.