Proposition 2 of this article: "Symmetry Groups of Platonic Solids" claims that the rotational symmetry group of a cube is isomorphic to $S_4$.
Note that there can be at most 24 rotational symmetries of a cube. It is possible to couple the opposite vertices of the cube, because if one of the vertices in the pair moves, the other pair would move correspondingly to remain an opposite vertex. Since there are four such pairs of vertices, the maximum rotational symmetries is 24, provided that all the permutation of these four pairs of vertices can be found. Indeed, inspection shows that this can be done.
We can indeed find 24 rotational symmetries acting on the four main diagonals. However, why does this imply that there are at most 24 rotational symmetries of a cube? Specifically, there are 6 faces/8 vertices/12 edges, why don't we try to find more rotational symmetries acting on them? Furthermore, why don't we try to find more rotational symmetries acting on some other objects embedded in a cube?

Consider the set of oriented edges of the cube, so that each of the edges counts twice, once in one direction and once in the other. There are 24 of them.
Now a symmetry of the cube maps oriented edges to oriented edges, and is complete determined to what it does to one of them. If follows at once that there are at most 24 symmetries.