The difference is what each method does with a negative or out-of-order argument.
slice(start, end)treats a negative index aslength + index, then clamps to[0, length](the spec'sToClampedIndex). Ifstart ≥ endthe result is"". So'JavaScript'.slice(-6)is"Script".substring(start, end)clamps each argument to[0, length]first, so negatives andNaNbecome0. Then it swaps them ifstart > end. So'JavaScript'.substring(-6)is the whole string, andsubstring(4, 1)is the same assubstring(1, 4).at(index)(ES2022) returns one code unit, counting negative indices from the end, andundefinedwhen out of range. ComparecharAt, which returns""out of range and doesn't accept negatives.substr(start, length)accepts a negative start, but it's an Annex B legacy method; useslice.
Why at was needed: str[-1] can't mean "last character". Bracket access is property lookup on any object, so -1 becomes the key "-1", which strings don't have. The relative-indexing proposal added at to strings, arrays and typed arrays instead.
The other everyday methods follow the same habits:
padStart(targetLength, padString)/padEnd(ES2017) repeat and truncatepadString(default a space) to reachtargetLength; a string already that long comes back unchanged.'7'.padStart(3, '0')is"007".replaceAll(search, replacement)(ES2021) replaces every occurrence of a string without needing a regex. With a regex, it must have thegflag, or it throws aTypeError.
Two facts apply to all of them: strings are immutable, so each returns a new string; and indices count UTF-16 code units, so slice and at can split an emoji's surrogate pair.