The following suggestions produce best results for Auto-Skeleton and are generally useful, regardless of how you set up the solve.

Actor stance

Marker placement

A- vs T-pose

Auto-Skeleton works with an A- or T-pose. The T-pose may give slightly better results because it assumes the clavicles are straight, whereas with an A-pose, it estimates the appropriate bend downward.

Single frame vs ROM

Auto-Skeleton looks only at the marker positions at the current frame. You need only a T- or A-pose to run it, but having a ROM present is useful to check whether the resulting calibration solves well.