Before you let Hermes build your idea, have it try to break the plan first. `/grill-me` takes the Grill Me concept popularized by Matt Pocock and pushes it further. Use it when you’re about to hand Hermes something bigger than a simple task: An app you want built. A workflow you want automated. A multi-agent system you’re designing. A migration or major change. A business process you want rebuilt. Or even a rough idea you know isn’t fully thought through yet. The original Grill Me idea is great because the AI keeps questioning the plan until the important decisions are exposed. Hermes adds another layer. It maps which decisions depend on others, then works through them in the right order. If a question depends on something you haven’t decided yet, it waits. If Hermes can answer something by researching your files, codebase, configuration, or available information, it’s supposed to investigate that itself instead of making you do the homework. And when it does need your judgment, it can give you its recommendation and explain why. So instead of: “Here are 15 things I need you to answer.” It becomes more like: “Here are the decisions we can resolve now. I researched what I could. Here’s what I recommend. Once we settle these, we can move to the decisions that depend on them.” It also deliberately looks for edge cases, failure modes, bad assumptions, rollback problems, and things you may have forgotten to define. Then it hands the resolved plan back before the actual build starts. That’s why I like this beyond coding. The bigger the job you’re about to hand an AI agent, the more valuable it becomes to make the agent challenge the plan before it starts executing it.