Grill Me Skill for Claude Code: How mattpocock's Plan Interrogation Works

Most planning failures happen because the agent started building before the developer finished thinking. grill-me, at 990,482 installs and 239,217 GitHub stars (skills.sh registry, GitHub), attacks that directly: it makes the agent relentlessly pressure-test a plan until every decision branch is resolved. Search interest for it is up 3,250% in the Skillselion trend index, making it one of the fastest-rising entries among the best skills for docs and planning.

Key takeaways

• grill-me by mattpocock records 990,482 installs and 239,217 GitHub stars (skills.sh registry, GitHub), with an A quality tier, LOW risk rating, and beginner complexity.
• It is a user-invoked slash command whose entire body is: run a grilling session. The heavy lifting lives in the sibling grilling skill (561,973 installs).
• Questions arrive one at a time, each with the agent's recommended answer attached, and the agent waits for feedback before continuing.
• Facts that can be found by exploring the codebase are looked up rather than asked; only genuine decisions reach the developer.
• The plan is not enacted until the developer confirms shared understanding, which makes it a validation gate rather than a generator.

What is the grill-me skill in Claude Code?

grill-me is a slash command that starts an interrogation session about whatever plan or design is on the table. The grill-me skill listing classifies it under the validate stage with scope as its subphase: its job is to stress a plan before any code exists. Its mechanism is deliberate: the agent walks each branch of the design tree, resolves dependencies between decisions one by one, and refuses to move forward while an answer is still open. The repository behind it, mattpocock/skills on GitHub, was last pushed on 2026-08-24.

How do you install and run grill-me?

One command installs it from the skills.sh registry:

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

Then type /grill-me after describing a plan, a feature, or an architecture decision in Claude Code. The session asks one question at a time, always with a recommended answer, because a wall of simultaneous questions produces shallow replies. Anthropic's agent skills engineering post explains why packaging a workflow this way beats pasting the same instructions into every session.

What is the difference between grill-me, grilling, and grill-with-docs?

They are three entry points into one interrogation engine. grilling (561,973 installs, skills.sh registry) is the model-invoked core skill that defines the interview behavior; the agent can trigger it on its own when a plan needs stress-testing. grill-me (990,482 installs) is the user-invoked command that starts that session on demand. grill-with-docs (843,921 installs) points the same technique at documentation, grounding each question in the docs you feed it; it has its own walkthrough in the grill-with-docs skill guide. Install all three; they share one repository and the agent picks the right entry point.

When should a developer reach for grill-me?

Reach for it when the cost of a wrong assumption is higher than the cost of twenty questions. Concretely: before committing to a database schema, before an API contract freezes, before a refactor that touches many files, and before handing a plan to a fleet of executor agents. It pairs naturally with review skills that work on the other side of implementation, like code-review by mattpocock (431,416 installs, skills.sh registry), covered in the best code review skills for Claude Code guide. For everything else this author ships, the Matt Pocock skills map charts the full set.

A complete starter stack

• grill-me (990,482 installs): the on-demand interrogation command.
• grilling (561,973 installs): the core interview engine the agent can invoke itself.
• grill-with-docs (843,921 installs): the same pressure applied through your documentation.
• code-review by mattpocock (431,416 installs, skills.sh registry): the post-implementation check on the code the plan produced.

Common pitfalls

• Answering on autopilot. The recommended answer attached to each question is a default, and accepting every default reproduces the agent's assumptions instead of testing them.
• Grilling too late. Once code exists, developers defend it; the skill sits in the validate stage because scope questions are cheapest before implementation.
• Confusing it with brainstorming. grill-me narrows a plan that already exists; idea generation is a different job with different skills.

A plan that survives a full grilling session is a plan two parties actually agree on, and that is the entire point. Browse the rest of the planning category on the docs and planning category hub.

Common questions