I kept auto-approving the plans my agent handed me — a wall of markdown built for an LLM, not a human. As the field moves from prompts to loops (Boris Cherny, Peter Steinberger), the plan is the last cheap moment to steer before an agent runs for hours. So I built plan-html: an interactive HTML plan that floats the goal and intent up front, shows flows and mocks visually, interviews you to extract what you actually need, sets boundaries and validation for your loops, and lets you discuss every part in bite-size pieces. It runs locally with a tunnel for phone review. Get it on GitHub: github.com/DoryZi/plan-html
