grill-me

Trigger phrases

Phrases that activate this skill when typed to Claude Code:

• grill me
• stress-test this plan
• interrogate this idea

What it does

grill-me is a Claude Code skill that turns Claude into a skeptical reviewer for a plan or idea. Instead of confirming what you say, it walks the decision tree one branch at a time, picks the most load-bearing assumption, and asks one focused question. You answer; it picks the next branch.

The output of a session is a plan with the hidden parts surfaced — the assumptions you hadn’t articulated, the steps you were going to “figure out as I go,” the contradictions between two adjacent decisions.

It’s not a replacement for thinking. It’s a forcing function that gets you to externalize what you’d otherwise leave fuzzy.

When to use it

Reach for it when:

• You’re about to spend more than a few hours on a plan and want to know if any branch is going to bite you
• You have an article angle or product idea that “sounds clean” but you suspect is thin in places you haven’t named
• You’re evaluating a feature or commitment and want a skeptical second pass that doesn’t flatter your conclusion
• The phrase “I’ll figure it out as I go” is creeping into your plan

When not to reach for it:

• Small, mechanical work where the path is obvious
• A plan you’ve already grilled today — diminishing returns are real
• Final steps where you’ve decided and just want to ship

Install

The skill is distributed via Pocock’s skills repo. Install via his recommended path (npx skills add or manual copy of the SKILL.md into .claude/skills/grill-me/) — see his repo README for the canonical install instructions.

Once installed, the skill activates on the trigger phrases above. No additional config required.

What a session looks like

A typical session has three phases:

1. You state the plan. Even a sloppy version is fine; the skill expects to refine it.
2. The skill picks one branch and asks one question. The branch it picks is the one with the most uncertainty given what you said. The question is narrow and specific — never “tell me more.”
3. You answer; it picks the next branch. After 5–10 rounds, the skill either surfaces a contradiction (which forces you to revise the plan), or it concludes the plan is sound and tells you why.

The discipline that makes it work: one question at a time, focused on the highest-uncertainty branch, no questioning of decided pieces. It interrogates the gaps, not the whole thing.

Receipts

Honest reporting on what grill-me is good for and where it falls short:

Where it works well:

• Catches unstated assumptions about user behavior, implementation paths, and downstream dependencies
• Forces explicit answers to “why this and not that” — the questions you’d otherwise hand-wave
• Exposes plans that look complete but rely on a vague middle step

Where it backfires:

• On a plan you’ve already locked in, the questioning feels obstructive — close it and ship
• It can over-interrogate small decisions; not every plan needs full grilling
• Quality of grilling tracks quality of the input. Vague plans get surface-level questions; specific ones get sharp questions

Pattern that works: trigger it on plans you’re about to invest more than a few hours in. The session takes ~5–10 minutes. Below that threshold, the overhead exceeds the value.

Source and attribution

Originally written by Matt Pocock. The canonical SKILL.md and supporting files live in the grill-me folder of his public skills repository.

License: MIT. You can install, adapt, and redistribute the skill, with attribution preserved.

This page documents the skill from a practitioner’s perspective — what it does in production use, where it shines, and where it doesn’t. For the formal spec and any updates, defer to the source repo.