AI column

Coding@donghyeok99Staff

Matt Pocock's skills: make your agent ask before it codes

Rebuilding his karting app with these skills, one developer got a 45-line domain doc and a design decision record from a 5-minute grilling

Type /grill-with-docs: questions Q1 to Q3 arrive with recommended answers, the design tree fills as you answer, and GLOSSARY.md and an ADR get written.
Type /grill-with-docs: questions Q1 to Q3 arrive with recommended answers, the design tree fills as you answer, and GLOSSARY.md and an ADR get written.
Who
Matt Pocock's daily skills, straight from his .agents directory, for developers building real apps with agents.
When
The repository opened on February 3, 2026; the latest release is v1.3.2 from October 29.
Where
A plugin for Claude Code, Codex and Copilot; Gemini CLI, Cursor, OpenCode, Windsurf and more via the skills installer.

What was hard

Ask an agent for a feature and it quickly builds something, but when you open it, it isn't what you meant. It doesn't know the project's words, so it uses twenty where one would do, and code written without tests only turns out wrong once you run it.

What it does about it

The repository keeps one small skill for each of those failures. The most used are /grill-me and /grill-with-docs: before writing code, the agent lays your plan out as a design tree, asks only the questions that can be settled now, numbered and with a recommended answer, and lets your answers decide the next round. It looks facts up itself and leaves you the decisions; settled terms go into GLOSSARY.md and hard-to-reverse decisions into an ADR as they happen.

The Skills logo at the top of the README
The Skills logo at the top of the README

What people made with it

blog.scottlogic.comColin Eberhardt2026-10-19

Rebuilt his karting PWA KartLog: a 5-minute /grill-with-docs gave a 45-line context file and an ADR, /to-tickets eight tickets; the tests, he notes, checked mocks more than code.

To put Matt’s Skills framework to the test, I decided to rebuild the app from scratch.
questionpro.comAkash Karyakarte2026-05-29

Ran /grill-me on per-workspace billing limits in a multi-tenant SaaS; the session exposed that he had never defined what users see when a workspace hits its limit.

Take a real feature — per-workspace billing limits in a multi-tenant SaaS.

How it runs inside

  1. 1

    You → Claude Code / Codex

    /setup-matt-pocock-skillsOnce per repo it asks for your issue tracker, triage labels and doc layout, and writes them under docs/agents/.skills/engineering/setup-matt-pocock-skills/SKILL.md
  2. 2

    You → Claude Code / Codex

    /grill-with-docsA one-line skill: it tells the agent to call two others, grilling and domain-modeling.skills/engineering/grill-with-docs/SKILL.md
  3. 3

    Claude Code / Codex → grilling

    The plan as a design treeEach decision hangs the decisions that depend on it; only the questions answerable now go into a round.skills/productivity/grilling/SKILL.md
  4. 4

    grilling → You

    Numbered Q1, Q2 … with a recommended answerWorded so that yes accepts the recommendation; facts in the code are looked up by a sub-agent, not asked.skills/productivity/grilling/SKILL.md
  5. 5

    Claude Code / Codex → domain-modeling

    Your terms checked against GLOSSARY.mdConflicting terms are called out at once, fuzzy ones get a precise name, and claims the code contradicts are raised.skills/engineering/domain-modeling/SKILL.md
  6. 6

    domain-modeling → GLOSSARY.md · docs/adr/

    Terms written at once, ADRs only when dueAn ADR is offered only when a decision is hard to reverse, surprising without context and a real trade-off.skills/engineering/domain-modeling/SKILL.md
  7. 7

    Claude Code / Codex → Spec on your issue tracker

    /to-spec: a spec labelled ready-for-agentProblem, solution, user stories and testing decisions, from the conversation and glossary, published to your tracker.skills/engineering/to-spec/SKILL.md

What goes in, what comes out

/grill-me + a paragraph of your planRounds of numbered questions with recommended answers, until every branch is visited.
/grill-with-docsThe same session, plus GLOSSARY.md and ADRs updated inline.
/to-ticketsA plan or spec split into tracer-bullet tickets, each declaring what blocks it.
/improve-codebase-architectureCandidates for deeper modules as an HTML report, then a grilling on the one you pick.

Is it for you

Good for

  • When what the agent builds often isn't what you wanted
  • When the agent spells out your project's jargon at length every time
  • When the agent pours out code without tests
  • When you'd rather pick and adapt small skills than hand the whole process to a framework

Not for

  • The engineering skills need /setup-matt-pocock-skills once per repo to know your tracker and doc layout.
  • /improve-codebase-architecture is a survey, not a rescue: the README says it won't untangle an old codebase for you.
  • Installs through the skills installer or Gemini CLI update by hand, and installing those alongside the plugin gives you every skill twice.
  • MIT licence: companies may change and ship it, keeping the copyright notice.

Reviewed by Devalley’s editor before publishing. How the column is made

Sign in to post, comment and like.Sign in

Report

Reason

Read the community rules