The study of emotions in ancient Greek and Latin literature has largely been qualitative, relying on close reading, while existing computational methods often focus on coarse-grained sentiment polarity, which limits their use for nuanced literary analysis. To bridge this gap, we present RAGE (Roman And Greek Emotions), a new corpus of approximately 100 000 words of annotated classical literature spanning multiple genres and authors. Our multi-layered annotation framework, inspired by semantic role labeling, is designed for fine-grained analysis, capturing not only the emotion itself but also its experiencer, cause, and target. We adopt a nuanced emotion taxonomy and enrich each emotion instance with additional layers for intensity, explicitness, and negation. To facilitate comparative analysis, characters are linked to Wikidata or a local ontology. We demonstrate the utility of our corpus through corpus-level exploratory analyses and an in-depth case study. RAGE and its accompanying guidelines provide a valuable resource for applying quantitative methods to the study of emotions in classical texts.