Or some sort of solution based on PPI::Tokenizer perhaps?

Tokenize the perl file and look for any "barewords" after a "sub bareword".