The positive mass theorem is more or less to do with the geometry of a type of positive scalar curvature condition.

Witten's work considers harmonic spinors, which are solutions to a certain linear elliptic system of partial differential equations. In his paper he presents a calculation which proves a rigidity theorem for harmonic spinors under a type of positive scalar curvature, directly comparable to Bochner's famous rigidity theorem for harmonic 1-forms in positive Ricci curvature. It is a little more complicated only since spinors are more complicated than differential forms.

Given the existence of a harmonic spinor with certain asymptotics, an integrated version of Witten's Bochner-type formula proves that the relevant positive scalar curvature condition implies the nonnegativity of the mass, now being expressed as the integral of the sum of squares of expressions built out of the harmonic spinor.

Witten's proof of the existence of such harmonic spinors is openly incomplete; he says "We have shown that the Dirac operator has no zero eigenvalue. Using this fact, we presume that standard methods can be used to yield" a key analytical step. The problem is existence of a Green's function for the relevant elliptic operator over noncompact spaces. Parker and Taubes gave a complete proof. I think it is not completely accurate to say that their proof only consists of "standard methods," since care is needed about weighted Sobolev spaces which can be somewhat delicate.

So I think it is inaccurate/misleading to put Witten's proof of positive energy theorem with some of his other works in terms of "inspiration and insight" for mathematics, or to just cite "origin in supergravity" (as Atiyah's laudatio does). His work here is a pretty direct and rigorous mathematical argument, in the vein of standard differential geometry. The gap is only due to his not being an expert in PDE methods. Even so he makes plausible that the relevant harmonic spinors exist. I think that a PDE expert reading his paper would likely even find it informally convincing.

As far as the Fields medal goes, I think some of his other work must have been more relevant. It's not hard to imagine someone else having discovered the main parts of Witten's proof, having to do with differential identities for spinors, with much less fanfare. The calculation relating the mass to the spinor is maybe the most striking part but I think many people (whether reasonably or not) would not regard it as a high point of mathematics (or whatever Fields medal is supposed to be about).