The most powerful line you can add to any AI prompt is three words: interview me first. Anytime you're building something new, a new skill, a new app, a new feature, end the prompt with "interview me to clarify any open questions." That one line forces a back and forth before a single piece of code gets written. You don't find out how bad your own instructions are until the AI starts making mistakes with them. The interview catches the blind spots before that happens, not after. I run this through an orchestration layer called Oh My Claude, one of the most powerful ones out there. A few commands sit on top of Claude Code: /interview runs a Socratic deep interview, targeted questions to fill any blind spots before it starts building anything. /plan runs consensus planning, multiple subagents, including Codex, actually review the plan before it gets approved. Not just one model rubber stamping itself. Next time you're building something brand new, try it. End the prompt with interview me first and see what it catches that you would have missed.