I had a look through the linked repo. One detail that may save you some time: it is a Java stored procedure with PL/SQL wrappers, not a PL/SQL BCrypt implementation.

There is an open issue from someone using it on Oracle 19c. BCRYPT_MATCHES returned ORA-00902 when called from SQL because the wrapper returns BOOLEAN. They reported getting 1/0 after changing the return type to NUMBER: MHeydari/BCrypt-algorithm-in-Oracle-plsql#1

The embedded BCrypt code also only accepts $2a$ hashes. $2b$ and $2y$ reach Invalid salt revision, so that is worth checking if you need to verify hashes created by another system.

I have not tested the full package on 19c myself, so I would not call it a confirmed solution. Are you mainly generating new hashes, or checking existing ones? If they already exist, which prefix do they use?