What Are Matt Pocock Skills? grill-me, TDD and 27 Skills: Contents, Installation and Use

You install Matt Pocock's skills; before starting work, your coding agent picks apart your plan with questions (grill-me), writes specs and tickets, develops test-first (TDD), sets up a debugging loop, scans the architecture, reviews the code and hands off the session. github.com/mattpocock/skills is an open-source set of 27 skills targeting four common problems when building real software with an agent (the agent doesn't understand what you want, talks more than needed, produces code that doesn't work, and makes the code messier over time). It's MIT-licensed and free; it installs into Claude Code from the official plugin marketplace, into Codex, GitHub Copilot and Gemini CLI with their own commands, and into other agents with npx skills.

What it does: Has the coding agent question your plan, test and review
Type: Skill collection (community)
Contents: 27 skills: 20 engineering, 7 productivity
Where: Claude Code, Codex, GitHub Copilot, Gemini CLI, Cursor and others
Version: 1.3.1 (04.10.2026)
Turkish: Not stated in the official documentation
Cost: Free, MIT

What do Matt Pocock Skills do?

A skill is an instruction file (SKILL.md) that tells an AI agent how to do a specific job. This set was written to solve the four problems the README lists: the agent misunderstands what you want, talks too long and vaguely, produces code that doesn't work, and complicates the codebase while moving fast. There's a skill for each problem: questioning, a shared glossary of terms, a test loop and architecture upkeep.

Have the agent question your planWhat you get: A plan clarified through Q&ANeeds a separate setup Test-driven developmentWhat you get: Small slices that pass their testsNeeds a separate setup Disciplined debuggingWhat you get: Root cause, fix and regression testNeeds a separate setup

Who it's for: people who write code with Claude Code, Codex, Cursor or Copilot and want the agent to work in a more controlled way. The README describes the skills as small, adaptable and composable; unlike approaches such as Spec-Kit that take over the whole process, it aims to leave control with you.

QuestionClarify the plan with /grill-me or /grill-with-docs.

WriteA spec with /to-spec, small tasks with /to-tickets.

Implement/implement does the work with tdd and closes it with code-review.

Wrap upImprove the setup and the code with /retro and /improve-codebase-architecture.

The flow is compiled from the skill descriptions in the README; each skill can also be used on its own.

Which skills are included?

The README splits the skills in two by who invokes them. The ones you invoke run only when you type /name and drive the work. The ones the agent can also invoke may be used by the agent on its own for a fitting task; they carry reusable discipline. The plugin (1.3.1) installs the 27 skills below.

Skill	Who invokes	When it helps
`setup-matt-pocock-skills`	You	Once per repo: choosing the issue tracker, triage labels and where docs go
`ask-matt`	You	Asking which skill or flow fits your situation
`grill-with-docs`	You	Plan questioning + updating the project glossary (GLOSSARY.md) and decision records (ADR)
`triage`	You	Moving issues through role-based states
`improve-codebase-architecture`	You	Scanning the codebase and seeing simplification opportunities in a visual HTML report
`to-spec`	You	Turning what you discussed into a spec and publishing it to the issue tracker (asks no new questions)
`to-tickets`	You	Splitting a plan or spec into small tickets with clear dependencies
`implement`	You	Doing the work in a spec or ticket; proceeding with tdd and closing with code-review
`implement-spec`	You	Implementing a whole spec on one branch with parallel sub-agents
`wayfinder`	You	Planning a big job that won't fit in one session with a map of decision tickets
`retro`	You	Getting suggestions to improve the agent's working environment after a session
`prototype`	Both	A quick answer to a design question with a single-file HTML page or a few UI variants
`diagnosing-bugs`	Both	Hard bugs and performance regressions: build a loop that goes red, narrow down, hypothesize, measure, fix, add a test
`research`	Both	Researching a question from reliable primary sources and having a sourced Markdown file written
`tdd`	Both	Developing a feature or fix test-first, slice by slice
`domain-modeling`	Both	Testing the project's terms against the glossary and stress-testing them with edge cases
`codebase-design`	Both	A shared language for "deep module" design: lots of work behind a small interface
`code-review`	Both	Reviewing a change on two axes: coding standards and conformance to the spec
`pr`	Both	Writing a pull request description with a summary, before/after evidence and a risk note
`wizard`	Both	Generating an interactive bash wizard for steps you have to do yourself (infrastructure, keys, dashboard settings)
`grill-me`	You	Having a plan or idea questioned until every open branch is closed (non-code work too)
`handoff`	You	Compressing the conversation into a handoff document another agent can continue from
`teach`	You	Learning a new topic over several sessions, using a folder as the workspace
`to-questionnaire`	You	Preparing a decision you can't make alone as a Markdown questionnaire for the person who will answer
`wait-what`	You	When you don't understand an agent message, having it re-explained in plain language with the missing context
`grilling`	Both	The questioning component used by grill-me, grill-with-docs, triage, wayfinder and the architecture skill
`writing-for-agents`	Both	Writing docs for agents: skills, AGENTS.md, CLAUDE.md

Your problem	Try this first
The agent misunderstands what I want	`grill-me` (non-code) or `grill-with-docs`
The agent gives code that doesn't work	`tdd`
I can't find a bug	`diagnosing-bugs`
The codebase got messy	`improve-codebase-architecture`
The session is full, I'll continue in a new one	`handoff`
I don't know which one to use	`ask-matt`

The repository also has misc (4) and in-progress (7) folders that aren't in the README reference; the plugin doesn't install them. The skills.sh page counts 55 skills including old names.

How do you install Matt Pocock Skills?

The README gives two ways: the plugin updates itself; npx skills copies editable files into your project and you do the updating. Pick only one per agent; if you install both, every skill loads twice.

Install from the official marketplace› claude plugin install mattpocock-skills@claude-plugins-official

Auto-update is on by default in this marketplace; to turn it off, use /plugin › Marketplaces (Claude Code docs).

Add the marketplace and install

› codex plugin marketplace add mattpocock/skills
codex plugin add mattpocock-skills@mattpocock

According to the README, it updates itself on startup.

Copilot CLI

› copilot plugin marketplace add mattpocock/skills
copilot plugin install mattpocock-skills@mattpocock

In VS Code, run the Chat: Install Plugin From Source command and enter https://github.com/mattpocock/skills. For auto-update in the CLI, the README describes adding a marketplace entry to the ~/.copilot/settings.json file once.

Install both folders

› gemini skills install https://github.com/mattpocock/skills.git --path skills/engineering
gemini skills install https://github.com/mattpocock/skills.git --path skills/productivity

To update, run both commands again.

npx skills (choose the agent with -a)› npx skills@latest add mattpocock/skills -a cursor

Supported examples: cursor, opencode, devin, windsurf, amp, pi; if you omit -a, it asks you to choose. When selecting skills, be sure to check setup-matt-pocock-skills. Update: npx skills@latest update; for new skills, run the add command again.

After installing, once per repo› /setup-matt-pocock-skills

What you get: the agent asks you which issue tracker you use (GitHub, GitLab, local files or another one you describe), which labels you give tickets, and where to save the docs it creates. Skills such as to-spec and triage use this setting.

How do you use them? Examples

You start the skills you invoke with /; the ones the agent can also invoke kick in by themselves on the relevant task. The output descriptions below are based on the README and SKILL.md texts.

Have your plan questioned"/grill-me I want to add a page where customers can book appointments and pay. Question my plan before we start."What you get: the agent doesn't write code; it asks you questions one by one until every branch of the design tree is clear (which payment method, cancellation rule, authentication…). The README recommends using it before every change; it has 1.5 million installs on skills.sh (October 29, 2026).

Find a hard bug"/diagnosing-bugs The payment form sometimes submits twice. Find the cause and fix it."What you get: a loop that progresses in stages: first it builds a feedback loop that reproduces the bug every time, then narrows it down, forms a hypothesis, measures, fixes it and adds a test so the bug doesn't come back.

Hand off the session"/handoff In the next session we'll finish the tests for the payment page."What you get: a handoff document with a summary of the conversation; it's written to the operating system's temp folder, not the project folder. The document suggests the skills the next agent should invoke; API keys, passwords and personal information are stripped out.

The skill texts are in English; you can write prompts in Turkish, but Turkish replies and glossary behavior are not stated in the official documentation.

What does it access, what does it write?

Reads

Project files, the git diff, GLOSSARY.md and decision records; issues in the issue tracker you choose; the research skill reads primary sources on the web.

Writes

Glossary, decision record, spec and ticket files in the project; to-spec and to-tickets publish to the issue tracker; implement writes code and makes commits.

Its own scripts

The skills work with the agent's own tools; the plugin has template scripts only for diagnosing-bugs and wizard. On the npx route, the installer (npx skills) runs on your computer.

Because the plugin updates itself, skill texts can change; read the SKILL.md before running skills that publish to the issue tracker or make commits in an important repo.

Is it free? License

The set is free and MIT-licensed. The only cost comes from the agent you use (a Claude Code, Codex or Copilot subscription, or API usage).

Allowed

MIT: you're free to use, modify and adapt it in your own projects and at your company
The README presents the skills as "change them to suit you"

Not allowed

No warranty (MIT)

Source: LICENSE and plugin.json. This is not legal advice.

How do you uninstall it?

Claude Code: Type /plugin, select mattpocock-skills in the Installed tab and choose Uninstall.
If you installed with npx: The skill folders were copied into your project, so delete those folders.
Codex, Copilot, Gemini CLI: Remove it from their own plugin or skill management; the README gives no uninstall command.

Frequently asked questions

What are Matt Pocock Skills?

An open-source set of 27 skills for coding agents: plan questioning, specs and tickets, test-driven development, debugging, architecture scanning, code review and session handoff.

What does the grill-me skill do?

Before the agent starts work, it questions your plan or design one question at a time until every unclear point is resolved. It's also used for non-code work.

What's the difference between grill-me and grill-with-docs?

grill-with-docs does the same questioning and also updates the project's glossary (GLOSSARY.md) and decision records. It's meant for code projects.

How do you install it in Claude Code?

With the command claude plugin install mattpocock-skills@claude-plugins-official. Then run /setup-matt-pocock-skills once in each repo.

How do you install it in Cursor?

With the command npx skills@latest add mattpocock/skills -a cursor. When selecting skills, check setup-matt-pocock-skills.

Can I install the plugin and the npx route together?

Not recommended. The README asks you to choose one route per agent; installing both loads every skill twice.

Are Matt Pocock Skills free?

Yes, MIT-licensed and free. The only cost comes from the coding agent you use.

Related guides

Sources

Not clear in the official sources: Quality with Turkish prompts · Uninstall command in Codex, Copilot and Gemini CLI (the README doesn't give one). Each is covered in its section.