The spec does not seem to define the origin for the texture coordinates.

One would then assume that the origin is bottom left like it is the case with OpenGL,
but looking at the Duck.gltf from the sample models it seem that it uses top left origin.

Texture coordinate origin needs to be defined.

If bottom left if chosen, then the sample models need to be fixed.

If top left is chosen, then it needs to be documented, that an implementation that renders
using OpenGL might need to flip texture coordinates or images.
(Or in the case of WebGL set gl.UNPACK_FLIP_Y_WEBGL to true.)