How to use mattpocock/skills with coding agents: setup and key skills

What a skill is and why use one

A skill is an instruction file that shapes how a coding agent approaches a task. The README describes the collection as the author's own agent skills, used every day for real engineering rather than open-ended 'vibe coding'.

The source says the skills are designed to be small, easy to adapt and composable, and that they work with any model. You can edit them, which is the main difference from a locked-down process.

• Use them when you want more predictable results than a single open-ended prompt gives you.
• Skip them if you want a full managed workflow; the README contrasts these skills with frameworks such as GSD, BMAD and Spec-Kit that own the process.

Install the skills for your coding agent

Pick one installation method per agent. The README warns that installing both a plugin and the copied files gives you every skill twice. Plugins update themselves, while copied files must be updated by hand.

• Claude Code: run the plugin install command. The README says it updates itself by default.
• Codex: add the marketplace, then add the plugin. It updates itself at startup.
• GitHub Copilot (CLI and VS Code): add the marketplace and install the plugin, then follow the README's extra settings step for auto-updates.
• Gemini CLI: install the productivity and engineering paths separately. Re-run both commands to update.
• Other agents or editable files: use the npx installer and pass an agent name such as cursor, opencode, devin, windsurf, amp or pi. Omit -a to choose interactively.

Run the setup command once per repo

Inside your agent, run /setup-matt-pocock-skills in each repository. The README says the setup asks three things, so the skills know how to work with your project. This step is required; the skills are not zero-config.

• Which issue tracker to use: GitHub, GitLab, local files, or anything else you describe.
• Which labels to apply when triaging tickets. The /triage skill uses labels.
• Where to save any docs the skills create.

Use /grill-me to align with the agent before coding

The README's first failure mode is misalignment: the agent builds something that does not match what you meant. The fix it proposes is a grilling session, where the agent asks you detailed questions about what you are building.

Use /grill-me for non-code work. Use /grill-with-docs for code changes, which the README describes as the same session with more added features. The author suggests using these every time you make a change.

• /grill-me: general questioning before a task.
• /grill-with-docs: the same questioning, plus shared-language and decision documentation (ADRs) as described in the README.

Build a shared glossary so the agent uses project language

The README argues that agents waste words decoding jargon. A shared language is a document that helps agents understand the terms used in your project. /grill-with-docs builds this with you and records hard-to-explain decisions in ADRs.

The README's example shows a glossary entry changing a long description into a short term, and says consistent naming also makes the codebase easier for the agent to navigate.

• Variables, functions and files get named consistently.
• The agent can navigate the codebase more easily.
• The author reports it spends fewer tokens on thinking.

Use /tdd and /diagnosing-bugs for tests and debugging

When the code does not work, the README says the agent needs feedback loops such as static types, browser access and automated tests. For tests, it recommends red-green-refactor: write a failing test, make it pass, then clean up.

The /tdd skill encourages this loop and gives guidance on good and bad tests. For bugs, /diagnosing-bugs wraps debugging practice into a loop that is gated phase by phase.

• /tdd: test-first, red-green-refactor.
• /diagnosing-bugs: a disciplined debugging loop with phase gates.