It greatly depends on how you want the modifications to behave, so you'll have to be more specific. In general, for a purely top down approach, all common references between subassemblies have a higher level skeleton.
Take, for instance, a car. You have a headlight module and a body module. Both would reference a higher level skeleton that defines the interface between the two. If the headlight needs to change in size, you change the top level skeleton and the changes flow from top to bottom.
More specifically, the top level skeleton might have an axis, point, datums, and/or surfaces that define the headlight location as well as features that represent the interfacing headlight size. The top level skeleton has publish geometries for those references and the headlight and body modules have copy geometry features that bring those references into the models. The models are built around these references and later assembled to the car assembly using a default constraint (unless it is something like a bolt that gets assembled to multiple locations). The headlight module can change in detail (more lightbulbs, LEDs, colors, etc.) as long as they stay within the scope of the skeleton reference to be sure that everything still fits with respect to other modules (i.e. the body module). The top level assembly will reflect the changes that occur within the module but if a scope change needs to occur, it happens at a higher level skeleton and flows down from there.
If you really need parameters and dimensions from a lower level passed upwards, then you'll need use a notebook to gather the references and link it to the assemblies, models, etc.
