This is a situation where I would argue that giving "exact" images of the output is not helpful, and I'm somebody who normally hates fruity presentations of info. Having the reader make their own associations allows it to be applicable across environments and develops what I'll go ahead and call "Rambo problem-solving", where you must get from point A to point B, even if that means you have to improvise or roll in the mud a little. You can't go off the beaten path with most GUIs, so being able to with a CLI is one of its most compelling uses. The reader needs to associate the command with the info, and not the presentation of it.
I never thought of that, but on reflection I think you're exactly right that hand-drawing is a good way to focus the reader's attention on the content of the text and not on the surrounding visual context. With a screenshot, the user might be consciously or unconsciously wondering why the terminal window looks different if it's a different OS/terminal emulator/font.
I feel that goes too far in the other extreme. They might not be able associate the examples with their own environment if the prompt is different (besides username and host; a different $PS1). Having the window representation makes for having more to associate, like "yes! I too have a window with a bit of text at the top and empty space below and to the right!" Having drawings also allowed for those arrows and text to explain the content, so that's great.
This is one of the reasons I'm a fan of Powershell, it's incredibly discoverable. Tab completed (with cycling) parameter names, names like Get-X/Set-X with consistent name mappings between them, pipelining with objects so there's no string parsing footguns, and Get-Member.
