There are three skeletons: a 65-bone humanoid, a 52-bone quadruped, and a 35-bone mech. Those counts are engine constants with drift-guard tests behind them, not marketing numbers.
Automatic rigging is very good at the parts that are geometrically obvious. Placing a spine, arms and legs on a character in a broadly standard pose is reliable. Weighting large smooth areas is reliable. If your mesh is clean and roughly A-pose or T-pose, most of the work lands.
It is weaker exactly where human riggers earn their fee. Shoulders and hips deform badly more often than anything else, because the correct answer depends on how the character will move rather than on the mesh alone. Faces, hands and anything with cloth or hair simulation are their own disciplines. Non-standard topology — fused limbs, extreme proportions, a mesh that is really two meshes — is where automatic weighting starts guessing.
The honest framing is that auto-rigging gets you to a riggable asset far faster than starting from nothing, and it does not replace a rigger for a hero character. Deformation rules run as part of validation precisely because a rig that looks fine in bind pose can be wrong the moment it moves.
- 65-bone humanoid, 52-bone quadruped, 35-bone mech.
- Deformation is validated, not assumed — a bind pose proves nothing.
- Clean topology and a neutral pose materially improve the result.