A proposed Claude Code workflow for AI classification with a model fallback

A user outlines a classification workflow to build in Claude Code: Fable creates a prompt and classifies 100 samples for spot-checking, while Jev handles initial classifications and sends low-confidence results to Luna. The user suggests having Fable evaluate the results, updating the prompt to catch edge cases and repeating the process. They claim this can produce a low-cost, highly accurate classifier.