Matt Pocock Skills

Status check 2026-11-13 — active

Verdict: The last commit was on 2026-11-12 (a merged PR on the software-factory skill), and releases v1.3.0 and v1.3.1 (2026-10-04) and v1.3.2 (2026-10-29) shipped within the last six weeks. Last release: v1.3.2 (2026-10-29). The CHANGELOG is managed with changesets, and the plugin auto-updates. · last commit: 2026-11-12 · licence: MIT Repo: https://github.com/mattpocock/skills Changed since this note: Since 2026-11-01 the only commits reword the beta software-factory skill in skills/in-progress (a glossary for designing a software factory, added 2026-10-31). The repo also has a skills/in-progress/ bucket the note doesn’t mention: chief-of-staff, claude-handoff, loop-me, setup-ts-deep-modules…

🚀 Description

mattpocock/skills — the agent skills Matt Pocock uses daily, framed as “real engineering, not vibe coding”. Explicitly the opposite of process-owning frameworks (GSD, BMAD, Spec-Kit): skills are small, adaptable and composable, work with any model, and leave you in control. ~60k people on the newsletter.

Install: claude plugin install mattpocock-skills@claude-plugins-official (auto-updates), Codex and Copilot via plugin marketplaces, Gemini CLI manually, or npx skills@latest add mattpocock/skills -a <agent> for editable copies. Then run /setup-matt-pocock-skills once per repo (issue tracker, triage labels, docs location).

🎨 The four failure modes it targets

  1. The agent didn’t do what I want → a grilling session (/grill-me, /grill-with-docs): the agent interviews you until every branch of the design tree is resolved. Use it before every change.
  2. The agent is too verbose → a shared language. /grill-with-docs builds a glossary (GLOSSARY.md) and ADRs inline. “The materialization cascade” replaces a sentence of jargon; naming gets consistent and thinking tokens drop.
  3. The code doesn’t work → feedback loops: types, browser access, tests. /tdd enforces red-green-refactor in vertical slices; /diagnosing-bugs is a gated loop (red-on-this-bug feedback loop → minimise → hypothesise → instrument → fix → regression test).
  4. We built a ball of mud → design every day. /to-spec quizzes you about the modules you touch; /improve-codebase-architecture surveys for “deepening opportunities” (Ousterhout’s deep modules) — run it every few days.

đź§© Skill catalogue

Split by who can invoke: user-invoked skills orchestrate and run only when you type them; model-invoked skills hold reusable discipline and the agent may reach for them.

  • Engineering, user-invoked: ask-matt (router), grill-with-docs, triage, improve-codebase-architecture, setup-matt-pocock-skills, to-spec, to-tickets (tracer-bullet tickets with blocking edges), implement, implement-spec (task graph across subagents on one integration branch), wayfinder (huge plans as a map of decision tickets), retro
  • Engineering, model-invoked: prototype, diagnosing-bugs, research, tdd, domain-modeling, codebase-design, code-review (two axes: Standards and Spec), pr (summary visual, before/after evidence, merge-danger call), wizard (interactive bash wizard for human-only steps)
  • Productivity: grill-me, handoff, teach, to-questionnaire, wait-what (user-invoked); grilling, writing-for-agents (model-invoked)

Reasoning for

I already run most of these through the plugin (the skill list in this session shows mattpocock-skills:grilling, tdd, diagnosing-bugs, code-review, pr, wizard…). The part worth stealing for my own skills is the user-invoked vs model-invoked split and the glossary habit — see Skill Authoring Best Practices and DOX — Self-Documenting AGENTS.md.

Alternatives considered

  • Superpowers — heavier, more opinionated end-to-end methodology (brainstorm → worktree → plan → TDD → review)
  • gstack and Karpathy Skills — other opinionated skill packs
  • Agent Skills (Addy Osmani) — engineering-lifecycle skills from a different angle
  • Vercel Skills — the npx skills installer used here

Template: tool