Matt Pocock Open‑Sources His Complete .claude Skills Repository
The article reviews Matt Pocock’s newly released mattpocock/skills GitHub repository, explaining its purpose, installation steps, folder structure, core engineering skills, four common failure modes, and how its concise, composable prompts differ from Anthropic’s official skills, while offering practical recommendations for Claude Code and Codex users.
Repository overview
The repository mattpocock/skills provides a collection of Claude Code skills that the author uses daily for real engineering work.
Installation
Install the skills with the community skills.sh installer: npx skills@latest add mattpocock/skills During installation you select the desired skills (the author selects all). The skills are placed under ~/.claude/skills/ or a project‑local .claude/ directory. Inside an agent run the bootstrap command: /setup-matt-pocock-skills The command prompts for three items:
Issue‑tracker choice (GitHub, Linear, or a local file).
Label taxonomy for triage.
Documentation directory.
After answering, the repository is wired to the project.
Repository layout
engineering/– core skills used daily for coding (the main set). productivity/ – generic, language‑agnostic workflows. misc/ – occasional utilities. personal/ – machine‑specific configs (not recommended for others). in‑progress/ – drafts under development. deprecated/ – retired skills.
Core engineering skills organized by failure modes
Failure mode #1 – Mis‑alignment
/grill-me(productivity) – forces the agent to interrogate the user and clarify each branch. /grill-with-docs (engineering) – same as /grill-me but also writes CONTEXT.md and ADR files.
Failure mode #2 – Verbosity (shared language)
Create a concise CONTEXT.md that captures project terminology. Example transformation:
BEFORE: “There’s a problem when a lesson inside a section of a course is made ‘real’.”
AFTER: “There’s a problem with the materialization cascade.”
The shorter phrasing reduces token usage for the agent.
Failure mode #3 – Code does not work (feedback loop)
/tdd– enforces a red‑green‑refactor cycle, requiring tests that target public behavior. /diagnose – six‑step debugging flow: Reproduce → Minimise → Hypothesise → Instrument → Fix → Regression‑test.
Failure mode #4 – Design degradation
/to-prd– converts a conversation into a PRD and opens a GitHub issue. /to-issues – splits a large plan into independent GitHub issues. /zoom-out – asks the agent to produce a high‑level module‑call graph. /improve-codebase-architecture – runs periodically (recommended every few days) to surface modules that could be refactored deeper.
Additional skills: /triage – routes issues via a state‑machine. /prototype – generates one‑off prototype code to answer design questions. /setup-matt-pocock-skills – the bootstrap script described above.
Productivity group (3 skills)
/caveman– removes articles and filler, reducing token usage by ~75 %. /grill-me – the interrogation skill listed under failure mode #1. /write-a-skill – meta‑skill that teaches the agent how to author a new skill.
Miscellaneous utilities
/git-guardrails-claude-code– intercepts dangerous commands such as git push --force, reset --hard, and clean before execution. /migrate-to-shoehorn – replaces as type assertions with the @total-typescript/shoehorn library. /setup-pre-commit – installs a full pre‑commit stack (Husky, lint‑staged, Prettier, type‑checking, tests). /scaffold-exercises – bulk‑generates exercise directories for course material.
Comparison with Anthropic official skills
Target audience is TypeScript/full‑stack engineers; many skills bind directly to the author’s migrate-to-shoehorn library and typical web‑backend workflows.
The “question the agent” pattern ( /grill-me, /grill-with-docs) forces clarification before action, mirroring senior‑engineer practice.
Each skill consists of a few hundred markdown lines, making them easy to modify.
The README cites classic engineering books (The Pragmatic Programmer, Domain‑Driven Design, XP, A Philosophy of Software Design) as theoretical foundations.
Key recommendations
/grill-with-docstogether with CONTEXT.md provides a language‑agnostic, high‑value workflow. /diagnose offers a reusable six‑step debugging SOP. /git-guardrails-claude-code adds a safety net for dangerous git commands. /caveman yields significant token savings.
Skills that assume GitHub or Linear (e.g., /to-issues, /to-prd, /triage) may need adaptation for alternative project‑management tools. Non‑TS‑specific utilities such as /migrate-to-shoehorn or /scaffold-exercises are optional.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
Old Zhang's AI Learning
AI practitioner specializing in large-model evaluation and on-premise deployment, agents, AI programming, Vibe Coding, general AI, and broader tech trends, with daily original technical articles.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
