I would first try to do calculations without ghost atoms, and then see how introduction of ghost atoms affects the affinity value. It is hard to say which position of ghost atoms is "good", without doing any calculations.

In general, LCAO basis set is tricky to use for affinity/work function calculations, especially, for open structures such as MoS2, i.e., not closed-packed structures such as Ag surface, for example, or other metal surfaces where position of ghost atoms is somewhat obvious. For accurate calculations, you may however try using plane-wave basis set (also available in QuantumATK) that does not require any ghost atoms to describe wave function tails in the vacuum far from the surface.