The configuration files read by freesasa_classifier_from_file() or the command-line option -c should have two sections: types: and atoms:, and optionally the section name:.
The types-section defines what types of atoms are available (aliphatic, aromatic, hydroxyl, ...), what the radius of that type is and what class a type belongs to ('polar' or 'apolar', case insensitive). The types are just shorthands to associate an atom with a given combination of class and radius. The user is free to define as many types and classes as necessary.
The atoms-section consists of triplets of residue-name, atom-name (as in the corresponding PDB entries) and type. A prototype file would be
The residue type ANY can be used for atoms that are the same in all or most residues (such as backbone atoms). If there is an exception for a given amino acid this can be overridden as is shown for PRO CB in the example.
A few example configurations are available in the directory share/. The configuration-file protor.config is a copy of the default classifier, and can be used to add extra atoms that need to be classified, while keeping the defaults for the standard residues (also see the file scripts/chemcomp2config.pl for instructions on how to generate configurations for new chemical components semi-automatically). If something common is missing in the default classifier, create an issue on Github so that it can be added.
FreeSASA also ships with some configuration-files that mimic other popular programs, such as NACCESS and DSSP.
The static classifiers in the API were generated using scripts/config2c.pl to convert the correspoding configurations in share to C code.