SkillAI coding

Matt Pocock’s AI coding skills, explained in plain English

Matt Pocock’s skills are a free, MIT-licensed set of small instruction files for AI coding agents such as Claude Code. Each does one job: question you before you build, write the test first, diagnose a bug in phases, or review a change. This page explains the main ones in plain words and says when to use each. It is my independent summary of his public repository, so treat the repository as the authority.

Go to the install commands
1

Copy the install commands

Nothing to fill in. Copy it as it is.

Install (from the repo README)7 lines
claude plugin install mattpocock-skills@claude-plugins-official # Other agents (from the same README):npx skills@latest add mattpocock/skills -a <agent> # Then, once per project, in your agent:/setup-matt-pocock-skills

The skills, in plain words

The repo’s README groups its advice around four problems with coding agents: the agent does not do what you meant, it is too wordy, the code does not work, and the codebase turns into a mess. The skills below map to those problems. The README also splits them in two: some you type yourself (for example /grill-me), and others the agent can reach for on its own when a task fits. Each description is my paraphrase of the skill’s own description and file.

  • Start here

    grill-me and grill-with-docs

    Use it when: Before any change bigger than a small fix, so you and the agent agree on what is being built. The README says to use it every time you want to make a change.

    The agent interviews you about a plan before any code is written. It asks in numbered rounds, suggests its own answer to each question so you can reply yes, and waits for your answers before the next round. grill-me suits plans that are not code. grill-with-docs does the same but also keeps a glossary file and records design decisions as you go.

    I would start with grill-me, or grill-with-docs if you are working inside a project folder. It adds no new tools. It makes the agent ask you better questions before it writes code. The README calls the grilling skills its most popular ones.

  • domain-modeling

    Use it when: When the agent uses long, vague wording or names the same thing in different ways. The README says agents dropped into a project often use 20 words where one will do.

    Builds the project’s shared vocabulary. It challenges loose terms, tests them against edge-case scenarios, and writes the agreed meanings into a GLOSSARY.md file and the decisions into records as you go.

  • tdd

    Use it when: When you build a feature or fix a bug and want steady, automatic feedback on the agent’s own code instead of trusting it to be right.

    Test-driven development in small loops. The agent writes one failing test first, then only enough code to make it pass, one small slice at a time. The skill tells it to agree where tests go before writing any, and to test behavior through public interfaces, not internal details. It leaves tidying up to the review step.

  • diagnosing-bugs

    Use it when: When something is broken, throwing errors, failing or slow. It also tells the agent to hide secrets in anything it shows you.

    A phased routine for hard bugs and slow code. The first and biggest phase is building a fast pass-or-fail check that fails because of this exact bug, and the skill says not to form theories until that check exists. After that it reproduces and shrinks the case, tests hypotheses, adds instrumentation, fixes the bug and adds a regression test.

  • code-review

    Use it when: Before you merge a branch or open a pull request.

    Reviews the changes since a point you name (a commit, branch or tag) on two axes: does the code follow the repo’s own standards, and does it do what the original issue or spec asked? The two checks run as separate helper agents with fresh context, so neither sways the other.

  • to-spec and to-tickets

    Use it when: When a chat has produced a plan and you want it written down in your issue tracker. Run the setup skill first so they know which tracker you use.

    to-spec turns the conversation you have already had into a spec and publishes it to your issue tracker, with no new interview. to-tickets splits a plan into small tickets, each a thin end-to-end slice that says which other tickets block it.

  • improve-codebase-architecture

    Use it when: Every so often. The README suggests running it every few days, and says it surfaces candidates rather than untangling a messy codebase for you.

    Scans a codebase for places where the design could be simpler (the repo calls these deepening opportunities), shows them in an HTML report, then talks through the one you pick.

  • research

    Use it when: When you want facts from official docs gathered while you keep working.

    Sends a background agent to investigate a question using primary sources, such as official docs and source code, and saves the findings as a Markdown file that cites the source of each claim.

  • prototype

    Use it when: When you want to test whether a logic or a screen idea feels right before building it properly.

    Builds throwaway code that answers a design question: either one shareable HTML file for a logic or state question, or several different UI variations you can switch between on one route.

  • ask-matt

    Use it when: When you cannot remember which skill to use.

    A router. Describe your situation and it points you to the skill or flow that fits, using the repo’s own map from idea to shipped code.

Source: Matt Pocock’s skills repository on GitHub (MIT license). Checked 2026-11-22, repository version 1.3.2.

How to use it

  1. 1Open the repository and read the README. It explains the four problems the skills target.
  2. 2Install it for your agent using the command from the README. It lists steps for Claude Code, Codex, GitHub Copilot, Gemini CLI and a general installer for other agents. Then run /setup-matt-pocock-skills once in each project. It asks which issue tracker you use, which labels you apply to tickets, and where to save docs.
  3. 3Start with one skill, such as grill-me or grill-with-docs before your next feature, and add others as you need them.
  4. 4Read each skill’s own file before you rely on it. They are plain text files, and the README invites you to adapt them.

A good result looks like

An agent that asks questions before it builds, tests its own work, debugs in steps and reviews before you merge. You stay in charge of the plan.

Stay safe

  • Skills are instructions an agent follows. Read a skill before installing it, as you would a script.
  • This page summarizes another author’s work. For accurate and current details, use the repository.

If it goes wrong

4 known problems, each with a fix

  1. 01

    Problem: Installing everything at once.

    Fix: Begin with one or two skills and add more when you feel the need.

  2. 02

    Problem: Skipping the setup step.

    Fix: Run the setup skill once per project so the other skills know your tracker and docs.

  3. 03

    Problem: Treating the skills as a fixed process.

    Fix: The repo says they are small and meant to be adapted, so edit them to fit your team.

  4. 04

    Problem: Installing both the plugin and the editable copy.

    Fix: The README says to pick one per agent, because installing both gives you every skill twice.

Has it been tested?

Not tested yetmodel: none yet · date: none yet

Not yet. This one was written for this page and I have not run it on a specific model yet. Results will vary by tool, so run it once, check the result, and use the fix table above as your checklist. When I test it, the model, the date and the result will be listed here.

Questions

Did you write these skills?

No. They are written by Matt Pocock and published under the MIT license. This page is my plain-English summary, checked against the repository on 2026-11-22.

Do I need Claude Code to use them?

No. The README lists install steps for several agents, including Codex, GitHub Copilot and Gemini CLI, and a general installer for others.

Which one should a beginner start with?

I would start with grill-me, or grill-with-docs if you are working inside a project folder. It adds no new tools. It makes the agent ask you better questions before it writes code. The README calls the grilling skills its most popular ones.

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

Both run the same interview. The README describes grill-me as the one for non-code uses, and grill-with-docs as the one that also builds your project’s glossary and records design decisions as you go.

Is it free to use?

Yes. The repository uses the MIT license, which lets you use, copy and change the files. Read the license in the repository for the exact terms.

Can you set up agents or automations like this for my team?

I build AI integrations, bots and automations. If you want an agent or an automation set up around your own process, describe it on the estimate page.

Build this for you

Want this built into your own workflow?

Tell me what you do by hand today. I will say what is safe to automate, what needs a person, and roughly what it would take. I reply within 24 hours.

Wondering what your project will cost?

Get a free price range and timeline in 24 to 48 hours. No commitment.

Get a Free Estimate