afni.nimh.nih.gov

You're right. The @auto_tlrc is for affine alignment, and auto_warp.py is for nonlinear alignment. The suggestion is to find a workaround. "skull_strip_input yes" means to skullstrip the input, so that could be the problem for you. You can skip the affine step with and use the output from @auto_tlrc if you like with "-affine_input_xmat ID" to apply an identity transformation (skip it) or supply the affine transformation output by @auto_tlrc.