Otto,
Even with a mask, a t-test on only 7 subjects is going to have a hard time finding a large enough set of significant voxels to give you a q-value in the range that you would like.
See this thread for similar advice.
As for the 3dClustSim... You need to estimate the smoothness of each of your subjects via 3dFWHMx and then average that smoothness value as your inputs for 3dClustSim. If you used afni_proc.py to generate your analyses with the -regress_est_blur_errts option, then the script will automatically generate a blur_est.${Subject}.1D for each subject that you can use.
Some recent threads of similar discussions:
[
afni.nimh.nih.gov]
[
afni.nimh.nih.gov]
[
afni.nimh.nih.gov]