mattpocock/skills — agent skills for engineers

What it claims

Thirty-odd Markdown agent skills, split into an engineering set and a general productivity set. The author built them to fix failure modes he sees in Claude Code, Codex and similar agents. Most of the summary's ground is already covered, but the page adds four things:

• An install choice that determines how updates work.
• A required per-repo setup step.
• A strict invocation rule between skills.
• Skills the summary doesn't name: ask-matt, wayfinder, retro, prototype, research, code-review, pr, wizard, handoff, teach, to-questionnaire, wait-what and writing-for-agents.

How it works

User-invoked skills are orchestrators and run only when you type them. Model-invoked skills hold the reusable discipline, and the agent can pick them up automatically. A user-invoked skill may call model-invoked skills but never another user-invoked one.

flowchart LR
  GM[/grill-me/] --> G[grilling]
  GWD[/grill-with-docs/] --> G
  TR[/triage/] --> G
  WF[/wayfinder/] --> G
  ICA[/improve-codebase-architecture/] --> G
  IMP[/implement/] --> TDD[tdd]
  IMP --> CR[code-review]

Some mechanics the summary leaves out:

• Domain docs: grill-with-docs and domain-modeling write GLOSSARY.md and ADRs inline. The convention was earlier named CONTEXT.md.
• implement-spec works one integration branch. It runs implementer subagents in parallel across the tickets that are ready, then closes with /code-review.
• code-review runs two parallel sub-agents so neither pollutes the other. One checks against the repo's standards plus a Fowler smell baseline. The other checks against the originating spec.
• to-tickets emits tracer-bullet tickets with blocking edges. They are written either to a local file or as native blocking links on a real tracker.
• diagnosing-bugs loops through these stages: reproduce red, minimise, hypothesise, instrument, fix, regression-test.
• to-spec does no interviewing. It only synthesises the existing conversation.

How to use it

Pick one install method per agent. Installing both a plugin and skills.sh gives you every skill twice.

• Plugin: updates itself.
• skills.sh: copies editable files into your project, and you update them by hand.

Claude Code (auto-updates):

claude plugin install mattpocock-skills@claude-plugins-official

Codex (updates at startup):

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

GitHub Copilot CLI:

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

Then add this once to ~/.copilot/settings.json:

{
  "extraKnownMarketplaces": {
    "mattpocock": { "source": { "source": "github", "repo": "mattpocock/skills" }, "autoUpdate": true }
  }
}

In VS Code, run "Chat: Install Plugin From Source" and enter https://github.com/mattpocock/skills. It updates daily.

Gemini CLI (re-run both commands to update):

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

Other agents (cursor, opencode, devin, windsurf, amp, pi):

npx skills@latest add mattpocock/skills -a <agent>

Omit -a to choose interactively. When the installer asks which skills to take, include setup-matt-pocock-skills. To update, run npx skills@latest update, and re-run add to get new skills.

Before using the engineering skills, run /setup-matt-pocock-skills once per repo. It asks for three things:

• Your issue tracker: GitHub, GitLab, local files, or anything else you describe.
• Your triage labels, which /triage depends on.
• Where to save generated docs.

If you're unsure where to start, /ask-matt routes you to the right skill. The author recommends running /improve-codebase-architecture every few days. It outputs a visual HTML report of deepening candidates.

Limits

• /improve-codebase-architecture is "a survey, not a rescue". On an old codebase it finds candidates but won't untangle them.
• skills.sh and Gemini installs don't update automatically.
• Several skills assume a configured issue tracker: to-spec, to-tickets, triage and wayfinder.
• The page shows no licence, no evaluation, and no measured effect on agent output. The claims rest on the author's experience.