Gained 19,679 stars in a single week.
mattpocock's skills repository, now at 91,699 stars, is the fastest-growing AI-related project on GitHub this week—surpassing the combined total of CloakBrowser's 9,007 and agentmemory's 7,103.
The content of this repository is actually quite straightforward: a set of skill files written for Claude Code. Each skill describes an engineering practice—how to conduct code reviews, how to write migration scripts, how to handle breaking changes.
In mattpocock's own words: "Skills for Real Engineers. Straight from my .claude directory."
In plain English: This is exactly what I use in my own projects, and I've open-sourced it.
This Isn't a Tutorial, It's "Encapsulated Engineering Intuition"
Many people treat these skills as tutorials. That's the wrong way to approach them.
A tutorial tells you "how to do it." A skill tells an Agent "how it should be done in this project."
The difference is: tutorials are for humans, skills are for Agents.
A tutorial would say: "You should run tests before submitting a PR to ensure there are no regressions."
A skill would say: "Automatically execute npm test before PR submission. If tests fail, block the merge and output a list of the failing test cases."
The former is advice; the latter is a rule.
The reason mattpocock's skills went viral is that he externalized the "correct way to handle this project" from a senior TypeScript engineer's mind into machine-readable rules.
A Trend That Makes Me Slightly Uneasy
I acknowledge that what mattpocock is doing is highly valuable. But I want to discuss a slightly uncomfortable angle.
Over the past decade, one of a developer's core competencies has been engineering intuition—that feeling of "I know how this project should be built." It comes from experience, past mistakes, and countless codebases reviewed. This kind of knowledge is hard to teach, hard to learn, and hard to replicate.
Now, someone has packaged this intuition into .md files, hosted them on GitHub, and made them available for anyone to download and use.
Once your Agent is equipped with mattpocock's skills, its code review quality might surpass that of a frontend engineer with two years of experience.
This isn't to say Agents are better than humans. It's to say: years of engineering experience accumulated by human engineers are now being replicated and distributed at an extremely low marginal cost.
Who benefits from this?
- Teams benefit: New hires + Agents + skills = rapid alignment with team engineering standards
- Companies benefit: The standardization of engineering practices improves dramatically
- mattpocock benefits: The influence brought by 91k stars is massive
What about everyday developers?
If your core competency is "I know how to do code reviews," and now a single skill file can achieve 80% of that result—where does your value lie?
But Don't Panic Yet
I'm not here to spread panic about "AI stealing developers' jobs." I want to highlight something else.
The value of mattpocock's skills doesn't come from the files themselves, but from mattpocock the person.
It's his decade of TypeScript experience, his deep understanding of type systems, and his teaching background at Total TypeScript that make these skills so well-crafted.
The skill files are merely a medium. The real value lies in the person behind them.
This means two things:
First, the ability to write skills is itself a new core competency. Those who can translate engineering practices into structured skill definitions will become increasingly sought after.
Second, the "unreplicable" nature of engineering intuition is diminishing, but the ability to "create new intuition" will never depreciate. Skills can replicate existing best practices, but they cannot invent undiscovered ones.
My Take
The 91k stars for mattpocock/skills is not an isolated phenomenon. It's a microcosm of the exploding "Agent skill ecosystem."
Addy Osmani's agent-skills has 40k+ stars.
academic-research-skills has 11k+ stars.
Anthropic is officially pushing the skills feature as well.
This trend is irreversible. Engineering knowledge is shifting from "tacit knowledge in human minds" to "explicit, distributable, composable, and executable rules."
This isn't a bad thing for developers, but it does mean the rules of the game have changed.
Previously, your value lay in what you knew. In the future, your value will lie in what kind of skills you can define.
What you know can be replicated. The ability to define skills cannot—because it requires continuous innovation and judgment.
So, instead of worrying about skill files replacing you, ask yourself: What engineering intuition of yours is worth encapsulating into a skill?
Primary Sources: