GeezerVison

String.prototype.endsWith always strigifies this value but explicit coercion wouldn't hurt

I think it would be a mistake to attempt to stringify a TypedArray, let's not bother with this warning if the template is not a string, wdyt?

Oh that's right, it can be native Uint8Array, too. Fixed that in the actual functions. 😅

As for warning, AFAICT there's no utf character that ends with 0x58 except for one-byte X, so Array.prototype.at.call(template, -1) === 0x58 should suffice for buffers.

I'm not sure it's an always correct assumption that the system would be using a UTF encoding set, but maybe it's fair to consider so as it's certainly very common 🤔 but probably folks who are using a Buffer/Uint8Array are doing so precisely so it can work on weird encoding sets? Maybe it's not worth overthinking it, it's just emitting a warning, so it's not like it would block anyone to have false positive.

The warning makes sense because the underlying implementation still might get confused.
glibc performs multibyte-unfriendly check with strspn, and it might be likely for mkdtemp on exotic platforms supporting prefixes longer than 6 to be as rough as for (int i = strlen(template); template[--i] == 'X'; template[i] = getRandomChar()) {}.
IMHO if someone writes precise code for specific platforms and charsets, it's probably up to them to simply ignore warning about template not being portable.