TL;DR
Matt Pocock released production-ready agent skills straight from his .claude directory. With 66k stars and 16k added this week alone, the community is paying attention.
Read next
The mattpocock/skills repository hit 62k stars on GitHub's weekly trending list with 17 composable Claude Code agent skills that tackle the four failure modes killing real engineering teams.
5 min readMatt Pocock's open-source skill pack for Claude Code gained over 5,500 stars in a single day - a curated set of agent skills built from real engineering work, explicitly rejecting vibe coding.
6 min readMatt Pocock made his personal .claude directory public and hit 857 new GitHub stars in a day. Here's what 18 production-tested Claude Code skills look like from a prolific TypeScript educator.
5 min readThe GitHub trending page for the past week has been dominated by skills packs and .claude directories. One repo stands apart from the rest: mattpocock/skills, described simply as "Skills for Real Engineers. Straight from my .claude directory." It closed the week at 66k stars with roughly 16,579 added over seven days - a velocity that places it in the same tier as major framework announcements. Matt Pocock is the creator of Total TypeScript, a widely respected TypeScript education platform, and his reputation as someone who synthesizes hard-won production knowledge drives this kind of reach. When he publishes a .claude directory, engineers actually read it.
The repo frames AI-assisted development around four recurring failure modes that most engineers have hit:
Misalignment - the agent builds the wrong thing confidently. The /grill-me and /grill-with-docs skills address this by requiring the agent to interrogate requirements before writing a single line of code. /grill-with-docs specifically anchors that questioning to domain language documentation you provide.
Verbosity - the agent over-communicates, wasting tokens and burying signal. The /caveman skill switches the agent into a stripped-down communication mode that cuts token usage by around 75 percent according to the repo documentation.
Code quality drift - feedback loops break down without static types, test coverage, and browser checks. The /tdd skill enforces a red-green-refactor loop. The /diagnose skill structures debugging into repeatable steps rather than free-form guessing.
Architecture decay - complexity accumulates across sessions until the codebase becomes hard to extend. The /improve-codebase-architecture skill runs a structured refactoring review to catch decay before it compounds.
Beyond those four, the repo ships a full catalog. The engineering tier includes /triage, /to-issues, /to-prd, /zoom-out, and /prototype. The productivity tier adds /write-a-skill so you can encode your own team norms using the same format. The miscellaneous tier covers git-guardrails-claude-code, migrate-to-shoehorn, scaffold-exercises, and setup-pre-commit.
A supporting convention runs through all of it: a CONTEXT.md file that establishes shared domain language across sessions. This is what prevents the agent from re-inventing terminology every time you open a new conversation.
Get the weekly deep dive
Tutorials on Claude Code, AI agents, and dev tools - delivered free every week.
From the archive
May 7, 2026 • 6 min read
May 7, 2026 • 9 min read
May 7, 2026 • 6 min read
May 7, 2026 • 8 min read
The installer is a single command:
npx skills@latest add mattpocock/skills
After that, open your agent and run the setup command to wire up issue tracking, triage labels, and documentation preferences:
/setup-matt-pocock-skills
From that point the slash commands are available in any session where your agent reads the .claude directory. There is no build step and no configuration file beyond CONTEXT.md. The skills themselves are markdown files, so you can read them, fork them, and adapt individual ones without touching the rest.
If you want to preview before installing, the repo is public and the skill files are readable directly on GitHub. The docs/ folder is not present here the way it is in addyosmani's agent-skills, but the README covers the core workflows clearly and the skill files are self-documenting.
The repo is most immediately useful to TypeScript and JavaScript engineers already working with Claude Code or Cursor daily. The install path assumes npx and the git-guardrails skill targets Claude Code explicitly, so that is the audience the author optimized for.
Beyond the tooling fit, the value proposition scales with team size and codebase age. If you are a solo developer on a greenfield project with clean architecture and a good test suite, you will get marginal value from the architecture and quality skills. The /grill-me and /caveman skills are useful at any scale.
Where the pack earns its place is in larger teams or long-running codebases where knowledge transfer is expensive and the agent consistently misreads context. The CONTEXT.md convention alone is worth adopting: it gives every agent session a shared vocabulary without requiring you to re-explain your domain in every prompt. The /to-prd and /to-issues skills convert discussion threads into structured requirements, which matters when multiple engineers are using the same agent configuration.
Engineers who have already built their own .claude directory will find /write-a-skill useful for formalizing and sharing what they already know.
The skills category is exactly the area skills.developersdigest.tech covers - a growing directory of agent skills, Claude Code plugins, and reusable .claude configurations for developers building with AI. The mattpocock/skills pack is a strong candidate to appear there.
What is worth noting for DevDigest readers is the contrast in philosophy between this repo and addyosmani/agent-skills (which we covered earlier this week). Osmani's pack is organized around the software development lifecycle - Define, Plan, Build, Verify, Review, Ship - and treats skills as phase gates. Pocock's pack is organized around failure modes and is intentionally smaller and more composable. Neither approach is wrong. They serve different mental models.
If your workflow is already structured around lifecycle phases, addyosmani/agent-skills maps cleanly. If you are troubleshooting a specific recurring problem - misalignment, verbosity, architecture rot - and want to drop in exactly one skill, Pocock's catalog is easier to pick from selectively.
Both represent a broader shift: senior engineering judgment is becoming a distributable artifact. The .claude directory is the new dotfiles.
Strengths: the pack is small, composable, and written by someone with a track record of translating complex engineering knowledge into accessible formats. The four-failure-modes framing is a better mental model than a feature list. The CONTEXT.md convention is a concrete workflow change, not a vague best practice.
Limitations worth naming: the skills are written for Claude Code and have not been independently verified on other runtimes, despite the installer being generic. The /caveman 75 percent token reduction claim is stated in documentation without a reproducible benchmark. Some skills like migrate-to-shoehorn reference internal tooling that may not map to your stack without adaptation. And as with any skills pack, the value you get scales with how consistently your team actually runs the setup step.
Worth trying. Not worth treating as a turnkey solution without reading the skill files first.
Technical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.
Turn a one-liner into a working Claude Code skill. From idea to installed in a minute.
View AppUnlock pro skills and share private collections with your team.
View AppCatch broken SKILL.md files in CI before they hit your team.
View AppConfigure model, tools, MCP, skills, memory, and scoping.
Claude CodePrefix prompts with ! to run shell commands directly, bypassing Claude.
Claude CodeExecute shell commands with persistent working directory in project bounds.
Claude Codeobra/superpowers picked up 1,600+ GitHub stars in a single day by solving the real problem with coding agents - they ski...
Matt Pocock's Claude Code skills library picked up 3,372 GitHub stars on a single day. Here's what's inside, how to inst...
Matt Pocock's .claude skills directory hit 74.7k GitHub stars, offering structured workflows that address the four most...
The mattpocock/skills repository hit 62k stars on GitHub's weekly trending list with 17 composable Claude Code agent ski...
obra/superpowers is a composable, opinionated methodology for AI coding agents - structured workflows covering planning,...
A single CLAUDE.md file inspired by Andrej Karpathy's observations about LLM coding mistakes gained 24,000 GitHub stars...

New tutorials, open-source projects, and deep dives on coding agents - delivered weekly.