I don't think you'd like the results if you had it auto-generate a bounding box from all the individual images.
It would include all the invisible parts of your skeleton and the collisions would not look good nor work intuitively and fairly.
Calculating this all the time incurs some performance cost if you have a lot of images, and it would keep changing the shape, which, depending on your system, would not behave correctly and possibly be updated at a wrong time step.
What you actually could do is just define one big box for your skeleton in Spine, then use that. No need for the many small boxes.
http://esotericsoftware.com/spine-bounding-boxes
Don't be confused with the terminology; the editor's "bounding box" doesn't mean an axis-aligned bounding box. It's actually just a solid 2D polygon.
You can add it on a slot on the root bone so it doesn't move. (as long as the root bone doesn't move). But you can add it on any other bone too.
At runtime, the type is a BoundingBoxAttachment (or spBoundingBoxAttachment in spine-c).