I've been working on this problem for my own language, and have landed on something more clear than just following a convention. Basically you use [] and () to specify if the left and right bounds are included or not (based off of interval notation: https://en.wikipedia.org/wiki/Interval_(mathematics)#Including_or_excluding_endpoints). e.g. for your case

--slice [1:5)    # include the left index. don't include the right index
--slice [1:5]    # include both left and right index
--slice (1:5]    # don't include the left index. include the right index
--slice (1:5)    # don't include the left or right index

potentially not relevant to your case, but my version supports an end keyword which you can do math on, similar to python's negative indexing

[2:end-3]    # start at index 2 (included) and go through till the third from last index (included)
(end-3:end]  # start at the third from last (excluded) and go to the end (included)

Personally I'm a fan of 0 indexing, but for your context, I think it would depend on how the user sees what they're slicing. E.g. if it was pages with page numbers, the numbers would indicate if it was 0 or 1 index based. If there's nothing to actually show the user, I think picking something reasonable and documenting it well is probably the best bet.

I think it would depend on the typical user base and how the rest of the cli operates. If it's typical array work or your users are typically programmers or otherwise know computing, then stick to 0 based indexing. If they're users of spreadsheets and rarely interface with zero-based indicies, then stick to what they know. Just document it well enough for everyone!

I'd also think inclusive is more intuitive. If they only want one element, then they can provide the single element, otherwise they get the full range.

Although, if your cli is trying to mimic another programming function. If it's very clear that's the intent, then follow the functionality of the parent function.

Anybody capable of using a CLI knows that the right answer is:

• index from 0
• end is exclusive.

Dijkstra points out why: https://www.cs.utexas.edu/~EWD/transcriptions/EWD08xx/EWD831.html

I know some programming languages use : for ranges and it is more legible if you support negative indices, but I think START-END is more natural reading and I’d use : for START:COUNT instead, e.g. 3:4 for 4 elements starting from 3, so elements 3,4,5,6 or 3-6.

You can even support both formats! (Feature creep warning)