Plan

Turns specs, briefs, or feature requests into structured task lists that can stand alone without the original context. It writes each task with enough background that someone reading it six months later, or an AI agent starting fresh, can actually execute it. The output lives in your docs folder or gets returned in chat, not pushed straight to your issue tracker. What's honest here is that it forces you to think through dependencies and acceptance criteria up front, which means you'll catch the "wait, how does error handling work across these three tasks" problems during planning instead of mid-implementation. Works best when your input is concrete enough to have real task boundaries.