Thanks Eric!
I was about to say that (plus a blank .frcmod file) is all you'd need for a case like this (where the residue:residue linkage is a simple, well-parameterised amide bond) - but that would run into the issue that the standard AMBER forcefield doesn't include any parameters for (standard atom):(GAFF atom) combinations so it'd fail on the protein backbone. I'm totally swamped at the moment so won't be able to experiment myself for a while, but Yang: if you want to try it out what
should work is to use your favourite text editor to manually replace all the GAFF atom types with the corresponding protein ones (you can find a table of GAFF atom types at
https://emleddin.github.io/comp-chem-website/AMBERguide-AMBER-atom-types.html and standard types at
https://www.uoxray.uoregon.edu/local/manuals/biosym/discovery/General/Forcefields/AMBER.html - use the all-atom carbons in the latter).
To convert to the format needed by ISOLDE, you'll need to first save each residue in isolation (which - Eric, correct me if I'm wrong - will require you to first split the structure using e.g. "split #1 atoms :XXY atoms :XXZ", followed by "save XXY.mol2 #1.1 gaff t; save XXZ.mol2 #1.2 gaff t") then create a couple of dummy .frcmod files (XXY.frcmod and XXZ.frcmod) in the same directory as the .mol2 files. I'm not sure of the minimal contents of a .frcmod, but you could start with the attached and see how that goes. To actually make the conversion, start ISOLDE, then, in the top ChimeraX menu, choose ISOLDE/Prep/Convert AMBER Files. If all goes well, that should create a .xml file for each .mol2/.frcmod pair it finds in the current working directory. In terms of converting over from GAFF to standard AMBER atom names, you can choose to do that either in the .mol2 or in the .xml files (personal preference, but I'd probably choose the latter since it doesn't have the fiddly layout requirements of .mol2).
Good luck... while I'd love to provide something much neater than this, at the moment I truly don't know when I'll find the necessary time.
Best,
Tristan