Fun to see this. We can achieve agentic behaviors with simpler prompts using today's models. Below are mine when I built Smallville in 2023 with GPT-4 base.