The grill-me Skill for Claude Code: Pressure-Test a Plan Before You Build It

The most expensive bugs are decisions nobody questioned. The grill-me skill by Matt Pocock exists to ask those questions before you write a line of code - relentlessly, one at a time, until the plan has no fuzzy edges left.

Key takeaways

• grill-me has 460,658 installs and 121,024 GitHub stars, and it is moving fast: 188 new installs in the last day (skills.sh registry, GitHub, 2026-07-06).
• It interviews you one question at a time about a plan or design, walking each branch of the decision tree until shared understanding is reached.
• Every question comes with a recommended answer, so the session drives toward decisions instead of stalling on open-ended prompts.
• When a question can be answered from the codebase, the agent explores the code instead of asking you - it only burns your attention on genuine judgment calls.
• It sits in the validate stage of the workflow: after an idea exists, before a spec is written. See the ranked validate-stage shortlist.

What is the grill-me skill?

grill-me turns the agent into a demanding technical interviewer for your own plan. You describe what you intend to build; it probes architecture, product intent, edge cases, and dependencies between choices until every branch is resolved. The one-question-at-a-time discipline is the core mechanic - it forces each decision into the open instead of letting a wall of questions get skimmed.

How do you install and trigger grill-me?

From the project root:

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

Then say "grill me" (or ask to stress-test a plan or design). The Claude Code skills docs cover how installed skills activate. Sessions work best when you paste or point at the actual plan - a PRD, an issue, a paragraph of intent - rather than describing it from memory.

What makes a grilling session effective?

Three behaviors distinguish it from just asking an agent "any questions?". It resolves dependencies between choices, so a database decision made in question three constrains the caching answer in question seven. It provides a recommended answer with every question, which keeps momentum and gives you something concrete to disagree with. And it reads the codebase first when the answer lives there, so you are never asked something grep could settle.

How does grill-me compare to brainstorming?

They bracket opposite ends of planning. brainstorming by obra (261,134 installs, skills.sh) shapes a vague idea into a design through guided dialogue; grill-me assumes a design already exists and attacks it until it holds up. Teams that use both typically brainstorm first, then get grilled on the result. If you only adopt one, pick the one matching where your plans usually fail - too vague, or too untested.

Who should use it?

Developers about to hand a plan to an agent for implementation get the most from it: ambiguity that survives planning becomes rework during execution. It also earns its place before writing specs, RFCs, or estimates. Browse adjacent picks in the productivity and planning category.

Common pitfalls

• Grilling a plan that does not exist yet - with nothing concrete to attack, the session wanders; brainstorm first.
• Answering from memory instead of letting the agent check the code - you will assert things the repo contradicts.
• Stopping at the first "good enough" - the value is in the branches you had not considered, which tend to surface late in the session.

Ten minutes of getting grilled is cheaper than two days of building the wrong thing.

Common questions