I’m not sure that I’m reading you correctly. Are you saying that you would rather it generated files, which the user would then load in code?
If so, I’m not sure that doing so would work very well: if I’m not much mistaken, the shader generator produces shaders based on the properties that it observes in the scene (such as whether normal maps are present). Since it doesn’t know the scene before it’s loaded, it’s not in a position to generate a file.
It could, perhaps, require a single run to observe the scene, after which the shaders could be loaded–but that seems cumbersome, and might call for another once-off “shader-generating” run if changes are made (or might even break the scene).