There is no way to tell on metrical grounds whether the syllable division was scul.ptus or sculp.tus since the first syllable would be closed either way. However, since pt does not occur word-initially in Latin, it was probably not a possible onset. Hence sculp.tus is a more likely syllable division. Similarly, one would divide sānctus ‘sacrosanct’ as sānc.tus, but spectrum ‘mirror’ and antrum ‘cave’ as spec.trum and an.trum.
The idea that those consonant clusters not permitted in initial position must be treated heterosyllabically in medial position goes back to antiquity. For example in An. Ox. IV 332 we read:
ὅσα σύμφωνα μὴ δύναται ἐν ἀρχῇ λέξεων ἐκφωνεῖσθαι, ταῦτα καὶ ἐν μέσῃ λέξει εὑρεθέντα χωρισθήσεται ἀλλήλων. (attributed to Herodian's Περὶ συντάξεως τῶν στοιχείων, Lentz (1870), p. 396, 1-2)
Such consonants as are unable to be pronounced at the beginning of a word are to be
separated from one another when they occur word-medially.
But, as a matter of fact, it's not clear that a necessary inference about the behavior of medial clusters can be drawn from the behavior or inventory of initial clusters. For example—as I learned from Adam Cooper of Cornell—Klamath, a Native American language spoken in Southern Oregon, has a wide variety of initial clusters but always splits medial CC clusters. But what about the other situation? Is it possible for a language to allow a richer set of medial onsets than initial onsets? In other words, does any language not permit e.g. initial kt but syllabify medial kt as an onset? This too is alleged to occur, but I haven't tracked down a case yet. In any event, the upshot is that we simply cannot be certain about the syllable boundaries in cases like sanctus and scultpus.