>> 		                ;; '(#x10A00 . #x10A5F)
>
> This last line should be removed, no?

Indeed – left over from an earlier experiment.

>
>> 		                '(#x10A3F . #x10A3F)
>>                         (list
>>                          (vector
>>                           (concat consonant
>>                                   "\\(?:" virama consonant "\\)*"
>>                                   modifier "*"
>>                                   virama "?"
>>                                   vowel "*"
>>                                   modifier "*")
>>                           1 'font-shape-gstring))))
>
> Note that according to the rule above, a sequence
>
>   consonant modifier vowel
>
> will not be composed, although it matches the regexp, because its
> second character is not a virama.  Is this okay?

Because of the '(#x10A3F . #x10A3F) bit? Yes, that may be a
problem. With Kharosthi Unicode, out of the following three
examples, the middle one (consonant + modifier + vowel) has its
vowel attached incorrectly

  𐨗𐨸𐨁 𐨣𐨸𐨁 𐨐𐨿𐨮𐨸𐨁

Cf. 𐨣𐨸𐨁 with modifier with 𐨣𐨁 without modifier: