We present a novel, scalable approach for coarse-grained semantic profiling of literary texts, intended as an alternative to existing explorative approaches like topic modeling. We propose using verb classes, essentially a coarse-grained variant of word sense disambiguation, to build interpretable semantic text profiles. In our approach, we classify verbal phrases based on the semantic class of their full verb, as formalized in GermaNet and WordNet. Our 10,000 manual annotations serve as a testbed for our retrieval-augmented generation (RAG) approach to classifying semantic verb classes. We test foundation models in multiple prompting setups with few-shot learning and RAG, illustrating that our approaches can almost, but not quite, match the performance of human annotators. Our application to four genre corpora (Romantic fiction, Regional literature, Crime fiction, Adventure fiction) of German-language texts demonstrates that such semantic profiles can reveal the thematic focuses of document collections. We release our corpus of more than 10,000 verb class annotations.