It would be much easier to just save the template with a layer mask instead if using a shape. Just put the shape in the layer mask. Then you don’t need any extra layer to clip to and you don’t need an action either. It is usually best to put the layer mask on the group. See the attached screenshot.
The reason why it won’t work with a clipped layer saved in the PSD is because SPA needs to throw away the existing player layer and create a new one when it does the face detection. So it will lose the clipping when it does that.
However, using a layer mask is a better way anyway.