The AI Enhancements section was interesting. I've been wondering about the best practices for agents interacting with long-running dev servers, and Astro 7's approach (run in background and have a logs command) seems like a good model.