When the number of categories is small and definitions are clear, using prompts to have the LLM perform end-to-end classification directly is reasonable, with low cost and fast deployment.