By linked in a hierarchical fashion I mean as a binary tree but where a child can only contain one parent this way any attribute of the parent node can cascade down to the child nodes.
In your case where you would require the need of a object to be linked to another object through the animation tree you would not include the animation tree into the scene graph but have a node which has a reference to the animation along with the model, any textures, special shaders and any other requirements which may effect child nodes. To create the grip object you would have the ability to reference a bone in the animation tree and then set a attribute to it, then any other object you create which has the grab attribute set to it. Once the object are running all you would have to do is set the grab object as the child of the grip object and if you have written the behavior to link the two together the two should lock together in the correct place.
I believe this is what L. Sprio was getting at with the scene graph being used to assist in situations where you need cascading data (though I could be miss interpreting what he/she said).