There are not many cases of Proto-Berber final *əv in Aujila (only two). These two have two different reflexes.
imp. sg. ĕrní pl.m. ĕrniyât pl.f. ĕrnîmet; pf. 1sg. ernîḫ 3sg.m. yernî; pf2. 1sg. ernîḫa 3sg.m. yernâya; aor. 1sg. a-rnîḫ 3sg.m. a-yérni; impf. 1sg. rennîḫ 3sg.m. irénni 3pl.m. rennân ‘to increase, add; to bid’
cf. Ghd. ărnəβ ‘to add’; Kb. ərnu ‘to add’; Zng. aṛīh ‘to increase’
The loss of *n in Zénaga is striking, but the other two cognates give enough basis to reconstruct:
imp. sg. arév (rarely: úrev) pf. 1sg. uréfḫ 3sg.m. yurév; 1sg. urífḫa, urívḫa 3sg.m. yurîva; aor. 1sg. a-urévḫ 3sg.m. a-yúrev; impf. 1sg. tārévḫ, tāréfḫ 3sg.m. itârev ptc. târęven ‘to write’
Berber, cf. Ghd. órəβ ‘to write’; Kb. aru ‘to write’; Zng. īrih, iʔrih ‘to dictate’
The *v in Zénaga reflects as vowel length, but vowel length and a glottal stop seem to be incompatible in this language resulting in two forms, one with metathesised *v and one with the glottal stop in place (Ghadamès o also points to an initial glottal stop).
So we have two cases of final *v with two different reflexes. They're different verb classes in most Berber languages, but that shouldn’t really affect the final consonant reflex.
I have no idea how to explain what exactly is going on here. But I thought it was interesting to point out.
Perhaps ĕrní was loaned from a nearby Berber Language with a reflex of *v, Siwi for example, although I haven’t found this word in Siwi. The word doesn’t look like that likely a candidate to borrow anyway.