The perturbation step of loop modeling occurs in centroid mode – a reduced sidechain representation. Reading between the lines a bit, this error almost certainly means there’s no valid centroid representation of your ligand (since you didn’t create one), so when the perturbation step tries to run it can’t make a centroid version of your protein and it crashes.
Here’s two things to try:
A) skip the perturbation step (“-remodel no” or something similar). This may reduce the sampling the protocol does, although I think with KIC it’s not really an issue.
Try adding your ligand to the centroid residue type set. Look for a flag called -extra_res_cen and pass it your ligand file. (Try passing it the fullatom ligand parameter file you already have and cross your fingers. If that doesn’t work, you’ll have to make a centroid parameter file, I have no idea how to do that.)