Longitudinal freesurfer recon-all mismatch between brain and brainmask

Hi,
I run recon-all command to create a template from multiple time-points, and the output was badly skull-stripped in the brainmask.mgz file, but the brain.mgz file was good and also the segmentations. The mask is required for further steps.

this is the code I used:

recon-all -base ${subject}_template -tp ${subject}_s1 -tp ${subject}_s2 -tp ${subject}_s3 -all

Hope you will have some insights.
Thanks,
Maya

brainmask.mgz:


and this is brain.mgz: