The /grill-me Skill

Posted on In ,

When an idea is important but still fuzzy, I often see teams jump straight into planning or implementation. The grill-me skill offers a better first move: challenge the idea until the hidden decisions are explicit.

What is the grill-me skill?

grill-me is a question-driven skill that helps you sharpen a rough idea into a decision-ready direction. It is created by Matt Pocock. Official documentation and usage examples can be found in the website.

The core behavior is simple:

  1. You bring an idea that is still vague.
  2. The skill asks focused questions in rounds.
  3. You answer, challenge, narrow, and refine.
  4. The session ends when there are no unresolved branches left.

Unlike code-generation workflows, this skill is about reasoning quality, not output volume.

Why this skill is different

Two properties make grill-me special:

  1. Stateless by design.
  2. Frontier-based questioning.

Stateless means it does not create files or persist project artifacts. You can use it with no repository, no setup, and no assumptions that your topic is even technical.

Frontier-based questioning means it asks only the questions that are valid given what has already been decided. This avoids random question chains and keeps the conversation coherent.

When to use it

Use grill-me early, especially when you have direction but not clarity.

Good use cases:

  1. Defining a new feature before writing a spec.
  2. Evaluating product or business options.
  3. Clarifying architecture choices before implementation.
  4. Stress-testing assumptions in writing or strategy.

If you can already describe scope, constraints, and trade-offs precisely, you probably do not need this step.

Installation script (global)

If you are using Codex or another agent that supports the skills.sh installer, use this command:

1
npx skills@latest add mattpocock/skills

In the interactive installer:

  1. Select only grill-me and grilling.
  2. Select your coding agent target.
  3. Choose the global install option (for all projects).

This pairing is important because grill-me delegates to grilling. Installing grill-me alone can fail to run as expected.

How to run a productive session

Most session quality comes from how actively you respond.

Recommended pattern:

  1. Start with one concrete idea and its objective.
  2. Push back when a question is too shallow or off-scope.
  3. Say “I don’t know” when uncertainty is real.
  4. Split oversized problems into smaller slices.
  5. End each round by summarizing what changed.

Passively agreeing with every question can produce long but low-value sessions. Active steering is what converts questions into decisions.

One question at a time mode

By default, grill-me often asks in rounds. If you prefer a slower pace, you can force one question at a time globally.

Add this line to your global CLAUDE.md file:

1
When grilling, ask one question at a time.

After adding it, start a fresh grill-me session and confirm the behavior by checking that the agent asks only one numbered question before waiting for your answer.

Grillable vs ungrillable questions

Not every question can be solved by discussion.

Grillable questions usually involve scope, constraints, priorities, or trade-offs.

Ungrillable questions usually require something concrete to react to, such as UI feel, copy tone, or interaction quality. For these, create a quick prototype first, then return with evidence.

A practical rule: if answers remain speculative after several exchanges, switch from talking to making.

Common pitfalls

  1. Session is too long.
  2. Scope is too broad.
  3. Questions degrade in quality over time.
  4. You leave with confidence but not commitment.

Mitigations:

  1. Break the problem into smaller decisions.
  2. Prefer rounds over endless single-question drift.
  3. Stop and prototype when discussion stalls.
  4. Verify that each major choice can be defended to someone new.

Where it fits in a larger workflow

grill-me works best as a pre-spec alignment step:

  1. Idea appears.
  2. Run grill-me to expose assumptions.
  3. Convert outcomes into a spec.
  4. Implement with fewer surprises.

This sequence reduces rework because you commit later, but with stronger context.

Final thoughts

grill-me is not a planning shortcut. It is a thinking-quality multiplier.

Use it when ambiguity is high and decisions matter. If you treat it as a collaborative interrogation instead of a passive interview, you get clearer scope, better trade-offs, and a much stronger starting point for execution.

Resources