Source: mattpocock/skills · Maintainer: community: Matt Pocock · License: MIT · Checked: 24 October 2026
What is the grilling skill?
The grilling skill makes your coding agent interview you about a plan, decision or idea, one round of numbered questions at a time, until every decision is settled and nothing is silently assumed. It is the shared interview engine behind several other skills in Matt Pocock's collection, including grill-me and grill-with-docs.
It lives in mattpocock/skills, which Matt Pocock describes in the README as "My agent skills that I use every day to do real engineering - not vibe coding." The skill's description reads: "Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases."
As of 24 October 2026, its skills.sh page listed 980.1K installs (snapshot, 24 October 2026).
What does it do?
The SKILL.md is short, and it rests on three ideas:
- A design tree. The agent maps your subject as decisions, each with further decisions hanging off it.
- The frontier. At any point, the frontier is every decision whose prerequisites are already settled: the questions the agent can ask now without guessing at answers it hasn't heard.
- Rounds. Each round asks the whole frontier at once. Each question is numbered and titled, followed by the agent's recommended answer on its own line, and worded so that "yes" accepts the recommendation. Then the agent waits.
Your answers settle decisions, the frontier moves outward, and the next round asks what those answers unblocked. A question that depends on another question still open in the same round has to wait for a later round. The grilling docs say thirteen questions typically land in about three rounds rather than thirteen.
The skill also splits facts from decisions. "Finding facts is your job, never the user's," it tells the agent. If a question needs a fact from the file system or a tool, the agent sends a sub-agent to look it up, and only the questions that depend on that answer wait. Decisions always go to you. The session ends when the frontier is empty, and the agent should not act until you confirm you share the same understanding.
Triggers. grilling has no disable-model-invocation line, so in Claude Code it is model-invoked: the agent can load it when you say "grill me" or ask to stress-test a plan, and you can type /grilling yourself. The README calls it "The reusable interview primitive behind grill-me, grill-with-docs, triage, wayfinder and improve-codebase-architecture."
Folder contents:
SKILL.md: the design tree, frontier, round format and the facts versus decisions rule.agents/openai.yaml: Codex display metadata, with the name "Grilling" and the short description "Stress-test thinking a round of questions at a time".
There are no scripts, templates or extra reference files.
How do you install the grilling skill?
The README offers two routes and says to pick one per agent, because installing both gives you every skill twice.
Claude Code plugin (whole collection, updates itself). The plugin is listed in Anthropic's official plugin marketplace:
claude plugin install mattpocock-skills@claude-plugins-official
Claude Code namespaces plugin skills as /plugin-name:skill-name, so it shows up as /mattpocock-skills:grilling.
Codex plugin. From the README:
codex plugin marketplace add mattpocock/skills
codex plugin add mattpocock-skills@mattpocock
Just this skill, as editable files. The command on skills.sh is:
npx skills add https://github.com/mattpocock/skills --skill grilling
The skills CLI accepts --skill more than once, so you can add --skill grill-me in the same command if you want the typed shortcut too. Add -g to install for all projects. For Claude Code, the folder lands in .claude/skills/grilling/ in your project, or ~/.claude/skills/grilling/ with -g, which are the standard Claude Code skill locations. Codex and Cursor read the same SKILL.md format; see Codex skills and Cursor skills.
claude.ai. Custom skills upload as a ZIP of the skill folder in Customize, then Skills, as the Claude Help Center describes.
How do you use it?
Describe what you want stress-tested and say "grill me", or type /grilling with your idea. Answer a round by number ("1 yes, 2 the second option, 3 no, here's why") instead of quoting the questions back.
Example prompts (suggestions, not quotes from the repo):
/grilling I want to replace our cron jobs with a queue. Grill me before I write anything.Grill me on this plan to split the monolith's billing code into its own service.Stress-test my idea of moving our docs site from a CMS to Markdown in the repo.I'm deciding whether to offer annual pricing. Grill me until the open questions are gone.
Requirements. No packages, CLIs or API keys. The sub-agent step works best in an agent that can dispatch sub-agents, such as Claude Code. The docs say model quality matters: lower-effort or non-frontier models tend to collapse the interview into a couple of questions and an outline.
If you are a skill author, the docs suggest calling /grilling from your own skill instead of writing a new interview.
When should you use grilling, grill-me or grill-with-docs?
All three run the same interview. The difference is how they start and what they write.
| Skill | Who can start it | Writes files | Best for |
|---|---|---|---|
grilling |
You or the agent | No | The plain interview, or calling it from another skill |
grill-me |
Only you | No | A typed command the agent never fires on its own |
grill-with-docs |
Only you | GLOSSARY.md and ADRs |
Plans inside a codebase, where terms and decisions should be recorded |
The docs also point to the repo's prototype skill for questions that talking cannot settle, such as how a screen should look or feel, and to wayfinder for efforts too big for one session.
Tips and limits
From the repo's docs, not from testing:
- The frontier is judgement, not a graph. The agent picks it by reasoning, so it can put two questions in one round and only later find that one answer should have changed the other. Tell it, and it reopens that branch next round.
- One question at a time is supported. Add
When grilling, ask one question at a time.to your global CLAUDE.md. The docs say readers who work in a second language or read slowly often prefer it. - No question cap. It is out of scope on purpose. Say "wrap up" in plain language, or break a big plan into smaller pieces.
- Watch for self-answering. If the agent answers its own decisions, the docs call that a bug in the run. It happens most when another skill runs grilling to resolve a ticket.
- Wrappers don't always load it. The docs report an open bug where grill-with-docs sometimes runs without loading grilling. The sign is every question at once with no recommendations; asking the agent whether it loaded grilling usually recovers the session.
- The yes/no wording was fixed. The docs page still mentions recommendations that argued against the question's wording. The current SKILL.md now tells the agent to word each question so "yes" accepts the recommendation, a change the repo lists as a pending patch (#706).
The skill runs no scripts and writes no files. It is MIT licensed, so you can copy and change it as long as you keep the copyright notice. The skills.sh audits showed Pass from Gen Agent Trust Hub, Socket and Snyk as of 24 October 2026.
Save it in your Kipt library
In Kipt, open Skills and paste the skill's GitHub folder link into the import box. Kipt reads the SKILL.md and its files; because this repo has many skills, it shows a list to pick from, so you can take grilling together with the wrappers that call it. You can also drop the folder or a .zip. Your library stays in your own private GitHub repo and installs into Claude Code or Codex with one command:
npx skills add https://github.com/<you>/<repo>.git --all -g
Kipt is not affiliated with Matt Pocock or mattpocock/skills.
Similar skills
- Grill me skill: the typed front door to grilling, for plans outside a codebase.
- Grill with docs skill: the same interview, plus a glossary and ADRs written as you go.
- Domain modeling skill: the vocabulary and decision records that grill-with-docs and triage pair with grilling.
- Prototype skill: for the questions an interview cannot answer.