Why the Top‑3 AI Skill Consists of Only a Few Sentences

What Is grill‑me?

A "Skill" is a set of instructions for an AI assistant. The grill‑me skill consists of a single file whose entire content is a short, structured prompt that repeatedly asks the user for details until a shared understanding is reached.

针对我的计划，反复追问每一个细节，直到我们形成共同理解。沿着决策树的每个分支往下走，逐一理清各项决策之间的依赖关系，有些选择必须等前一个问题确定了才能回答，AI 会按这个先后顺序逐个问清楚。每个问题都要给出你的推荐答案。 每次只问一个问题，等我回答了再问下一个。一次抛出一堆问题会让人不知所措。 能通过查看环境（文件系统、工具等）找到的事实，直接去查，不用问我。但决策是我来做的，每个决策都要等我拍板。 确认双方理解一致之前，不要开始行动。

Hands‑On Demo

Installation is a single command:

npx skills add https://github.com/mattpocock/skills --skill grill-me

After adding the skill, the author opens an AI coding tool (Cursor) and invokes /grill-me with a vague requirement. The AI then proceeds through a series of interactions:

It asks for clarification of the goal and presents its own suggestion; the user simply confirms.

It queries the desired output format (Web app, CLI, etc.) and the user selects option 3 (Web app).

It lists required features; the user greedily requests all of them.

It proposes a technical stack (Electron for desktop) and the user accepts.

It confirms detailed choices such as framework, component library, and data storage, following the decision‑tree logic.

When the AI lacks a preference, the author follows the AI’s recommendation.

Throughout, the user frequently replies with “accept your suggestion” to keep the dialogue flowing.

After more than ten rounds of Q&A, the AI produces a complete project plan, which the author confirms. The AI then generates the code, resulting in a functional desktop skill manager in roughly ten minutes.

Why grill‑me Works

The skill targets the most critical stage of AI‑assisted programming: clarifying requirements before code generation. Most developers give AI vague prompts, leading to guesswork and rework. grill‑me forces the user to articulate every decision, reducing ambiguity.

Its questioning follows a decision‑tree pattern—each answer unlocks the next branch—ensuring no critical detail is missed. By limiting each turn to a single question, it avoids overwhelming the user.

The human‑AI division is explicit: the AI gathers factual information from the project environment, while the user makes only the strategic decisions. The optional “provide a recommended answer” clause lets the AI suggest a choice, and the user can accept it with a simple “yes,” streamlining the process.

The author likens this to the “rubber‑duck debugging” technique, but with an AI that can both interrogate and fetch information, turning the practice into an executable workflow.

Broader Implications

Beyond AI coding, the same pattern can guide any decision‑making scenario—learning paths, hardware selection, health plans—by letting the AI ask structured, incremental questions and the user provide clear answers.

Repeated use of grill‑me helps users think more clearly and produces more reliable outcomes.