Really enjoyed this! Thank you for sharing your approach and notes on process.

This framework makes a lot of sense. I think the general approach of eschewing one-shot prompts in favor of letting the LLM craft its own prompts (per your standards) is the missing piece in a lot of workflows.