Why Clarifying Requirements First Boosts AI Agent Success: The grill‑me Skill in Action

Problem: AI Agents Act on Incomplete Requirements

When users give a high‑level request such as “Help me redesign the homepage,” the agent often proceeds immediately, filling in missing details with its own assumptions. This can lead to decisions—layout, module order, conversion paths—that are not validated by the user, causing costly rework.

Introducing grill-me

grill-me

is a skill from Matt Pocock’s skills repository that changes the agent’s workflow: before any execution, it repeatedly asks the user key questions until both parties reach a clear, executable consensus.

Core Mechanism: Interactive Requirement Convergence

模糊需求
 ↓
上下文读取
 ↓
关键决策识别
 ↓
逐项追问
 ↓
用户确认 / 修正
 ↓
形成共识
 ↓
进入执行

This transforms the agent from “directly generate” to “clarify first, then execute.”

Design Principle 1 – Ask One Question at a Time

The skill uses a conversational decision tree instead of dumping a long questionnaire. It selects the next most critical question based on the current context and the user’s previous answer, reducing cognitive load and allowing later questions to depend on earlier answers (e.g., if the homepage goal is blog reading, subsequent questions focus on article display and subscription placement).

Design Principle 2 – Provide a Recommended Answer

Each question includes a default suggestion. For example, instead of asking simply “What is the core goal of the homepage redesign?” it asks “What action do you most want visitors to take—read the latest article, learn about you, or convert to a course? Based on the current site, I recommend prioritising blog reading.” The user can confirm or correct the suggestion, keeping the dialogue efficient.

Design Principle 3 – Separate Fact Retrieval from Decision Queries

grill-me

distinguishes between factual information that the agent can retrieve (e.g., existing components, current routing) and decision information that requires user confirmation (e.g., which goal to prioritize). This avoids unnecessary user prompts for facts while ensuring critical decisions are not made autonomously.

事实信息：Agent 应自行检索
决策信息：Agent 应请求用户确认

Case Study: Personal‑Website Homepage Redesign

Original request: “Help me redesign the homepage.” Without clarification, the agent might arbitrarily change layout, module order, and visual focus. grill-me first asks, “What action do you most want visitors to complete? I suggest blog reading.” After the user confirms, it proceeds to ask about module retention, subscription placement, and visual hierarchy, eventually producing a consensus list such as:

首页以博客内容为主
订阅入口需要前置并重复出现
主视觉不再直接使用最新文章，而是固定展示最新长文
非核心模块降级展示
旧的个人介绍和辅助入口保留但降低权重

Only after this agreement does the agent move to the coding phase, dramatically reducing the chance of mis‑aligned implementation.

Beyond Code: General Decision‑Clarification Tool

The skill is applicable to any task that requires clear trade‑offs before execution, such as topic selection, course design, product feature definition, or project prioritisation. By breaking a vague request into a series of decision points, the user receives a reusable set of clarified choices.

Comparison with Superpowers

Both enhance AI Agent workflows, but grill-me focuses on a single, pre‑execution clarification gate, while Superpowers provides a full‑stack workflow system that orchestrates multiple abilities throughout the entire development cycle. grill-me is triggered manually (e.g., via /grill-me) and stops the agent until the user answers; Superpowers automatically suggests capabilities as the conversation progresses.

grill-me：用户说“现在来拷问我”
superpowers：系统说“当前任务可能需要这些能力”

Thus, for a clearly scoped but ambiguous task, use grill-me; for a long, complex engineering effort, adopt Superpowers or combine both.

Why It Cuts Rework Cost

Rework usually stems from wrong upstream decisions rather than implementation bugs. By moving costly decision validation to a low‑cost dialogue before any code is written, the skill turns expensive post‑implementation fixes into cheap pre‑execution confirmations.

执行前澄清：一次回答即可修正方向
执行后返工：需要重读、重写、重构或回滚

Suitable Scenarios

Any task where the goal, scope, or trade‑offs are unclear but have a large impact, such as:

Redesigning an existing page

Designing a new feature

Refactoring core code

Writing a technical article for a specific audience

Planning a course outline

Generating product or technical proposals

Prioritising project work

Even simple tasks may not need the skill, but whenever decision points exist, a clarification step is valuable.

Related Tools

grill-me

handles only the questioning phase. grill-with-docs records terminology and decisions for long‑term projects, while batch-grill-me presents a batch of questions at once for faster information gathering.

grill-me：轻量追问，适合对话式澄清
grill-with-docs：追问 + 文档沉淀，适合长期项目
batch-grill-me：批量问题，适合快速收集上下文

Implications for AI Agent Workflows

The popularity of grill-me highlights a shift: AI Agent performance depends not only on model capability but also on the surrounding execution process. A disciplined workflow—clarify, plan, execute, verify—prevents agents from jumping straight to generation and reduces costly mis‑alignments.

Do not let an Agent jump from a vague requirement to implementation; first identify and close key decision points.

Conclusion

grill-me

redefines the starting point of AI Agent collaboration. By asking one question at a time, offering suggested answers, and separating factual retrieval from decision making, it turns ambiguous requests into concrete, executable consensus, lowering rework for code, writing, and product planning alike.