Maybe I need to take a step back. If, and heavy on the if, you want to communicate issues with humans, you're doing it wrong. If you want to rely on LLMs (and look where that's getting you), by all means do it that way. With humans, they need to come to that conclusion on their own. An example is like this:

You: "I would like to remove blurry frames with ffmpeg. I think I want to use xyz method. Can anybody please help me craft the command?"

Them: "yes, here's the command" (and maybe it's exactly what the LLM output)

You: "it's giving this error code. Here's a snippet of my input file. Can you please check if that works on your end?"

Them: "oh wow you're absolutely right, this doesn't work, I'll put in the issue ticket"

This is a lot nicer than "something else told me how to use your tool and it's broken".