GitHub Trending · Oct 30: Agent Skill Libraries and AI Browsers Take Off

GitHub Trending has a bunch of new faces today — yesterday none of them were even there.

You're not wrong, and this time every single new entry is tied to the agent ecosystem.

Let's go one by one. First up, today's biggest mover: mattpocock/skills.

It gained over 1,700 stars in a day, hit 164K total, and shot straight to number six.

The key is its positioning — the title literally says Skills for Real Engineers.

And they flat-out said it: this was pulled straight from my own .claude directory.

Wait — so the author open-sourced all his private prompts and skills?

Exactly, that's why it blew up today. Real, direct, and a little bit flexing.

The point is it fills a gap: engineers don't know how to write high-quality skills.

So it's basically a ready-made answer key — just copy the homework.

Right — perfect for people who use Claude to code every day but write half-hearted skills.

Okay, number two, right behind it: anthropics/skills. The official one showed up too.

This one's worth more attention. 167K total stars, a bit higher than the last one, up 400-plus today.

The official skill repo versus the community personal version — that's a head-on collision, isn't it?

Here's a key detail: the official one is written in Python, the community one in Shell.

Official is more like infrastructure; the community one is more like a field manual. They complement each other.

So which one do you think people should look at first?

Want to learn the basics? Go official. Want to quickly copy homework? Go with mattpocock's.

Number three: pingdotgg/t3code. Stars aren't climbing as fast, but it's interesting.

Yeah, 24K-plus stars, up 250 today, written in TypeScript.

You probably remember the pingdotgg team — the folks behind T3 Stack.

So they went from scaffolding all the way to AI coding tools?

That's not the point. The point is it very likely follows that type-safety philosophy.

So for frontend teams wanting to hook into agents, this is a pretty solid starting point.

Yep, good for teams already on T3 stack — low migration cost.

Then number four: lightpanda-io/browser. This one surprised me the most.

Why surprised? It's written in Zig — the language itself is niche enough.

37K-plus stars, up 500-plus today. A headless browser built specifically for AI and automation.

Isn't that stepping right into Puppeteer and Playwright's territory?

You could see it that way. But its selling point is lightweight — built for agents, without Chrome's bulk.

Think about it: if an agent has to run dozens of web tasks, memory and startup speed are make-or-break.

So they rebuilt the browser from scratch for agents. That's a bold move.

To sum up, its positioning is a low-cost browser foundation built specifically for agents.

So if you put all four together, do you see a common thread?

It's obvious. They're all agent ecosystem components — skill libraries plus a runtime foundation.

People aren't competing on models anymore — they're competing on how to feed and stabilize agents.

Right. Skills are the knowledge layer, browsers and code tools are the execution layer. The division of labor is getting clearer.

And two skill repos hitting the charts the same day shows the concept of 'skill' has really been accepted.

It's in the same lineage as those memory projects you covered before — agents are becoming engineered.

Alright, that's it for today. My must-see pick is lightpanda — it's the most counterintuitive.

I'll add anthropics/skills — the official catalog is worth a quick scan first.

Sounds good. That's it for now — see you on tomorrow's chart.