I Curated 800+ Agent Skills So You Don't Have To

Every agent framework ships a "skills" feature now — Claude Code skills, the open Agent Skills format, and whatever your harness calls them. The problem isn't that skills exist. It's that good skills are scattered across dozens of repos, and the ones you actually need keep resurfacing under different names in different places.

So I stopped hunting and started curating. I built a single index that collects agent skills from around the web and files them into clear categories, so you can find the right one in ten seconds instead of opening a new tab per skill.

This post is the full tour: what a skill actually is, why the hub exists, and every skill in it, category by category, each linking back to its source. The hub is on GitHub, open right now — you can browse it and have a working skill installed in about two minutes.

What "agent skills" actually are

At the simplest level, a skill is a folder containing one markdown file, SKILL.md, that an agent reads when it needs help with a task. It's context you hand a model on demand: how to write Swift the modern way, how to audit animations before touching code, how to delegate work to sub-agents.

A skill folder can bundle scripts, reference docs and templates alongside that file, but SKILL.md is the required part — metadata (a name and a description) plus the instructions themselves.

The ones worth keeping aren't generic "be helpful" reminders. They're replicable, high-leverage workflows — instructions that encode how a skilled person actually approaches a slice of work. A good skill turns "I hope the model figures this out" into "here's exactly how to do X well."

That's the bar I used when curating: does this skill encode a repeatable workflow, or is it just vibes?

Why they load on demand

Skills work through progressive disclosure, in three stages:

1. Discovery — at startup the agent sees only each skill's name and description, just enough to know when it might be relevant.
2. Activation — when a task matches a skill's description, the agent reads the full SKILL.md into context.
3. Execution — the agent follows the instructions, running any bundled scripts or reference files as needed.

That's what makes skills cheap. You can keep hundreds on hand and only pay context for the few that apply. It's also why a hub matters: an agent can only pick from skills you've installed, so the quality of your skill folder is the ceiling on your agent's workflow quality.

Why I built the hub

I'd seen the same great skills pop up repeatedly in different repos:

• An animation skill that asks "should this even animate, and what curve makes it feel right?"
• A "write Swift like it's 2026" skill covering concurrency, generics, and macros.
• A reading-only skill that scans a codebase for animation opportunities and rejects everything that shouldn't move.

Each time, I'd rediscover it, reconstruct the mental model behind it, and wonder why something this useful wasn't easy to find. The skills themselves weren't the hard part. Finding them was.

So I built a hub. The goal wasn't to own the skills — that would just add another copy to the sprawl. It was to be the place you go to discover what already exists.

How it works: an index, not a copy

This is the part I'm proudest of. The hub is an index, not an archive. Each skill is recorded as a linked entry under its category, pointing back to the source repo. We don't clone repos in — links keep the hub small and always current with the original.

The tradeoff is intentional. A copy goes stale the moment someone pushes a fix upstream. A link is a promise that you're always looking at the current version. For a collection this wide, staleness would defeat the whole point.

skills/
├── ui-ux-design/        # UI/UX & Frontend Design
├── motion-animation/    # Motion & Animation
├── video-motion/        # Video & Motion Video
├── coding-language/     # Coding & Language Guidance
├── coding-harness/      # Coding Harness & Agent Orchestration
├── content-media/       # Content & Media Generation
├── research-web/        # Research & Web
└── general/             # General / Meta Skills

Eight categories. One LINKS.md per category. Each row names the skill, links to the source, and says in one line what it's good for.

90 entries live in those eight files today, plus a combined skill index in the README. Below is all of it.

How to use the hub directly

The hub is on GitHub at github.com/YsYeoh/agent-skills-hub. Three ways in, fastest first:

• Browse a category file. Every LINKS.md renders as a table straight in the browser, no clone needed — for example Research & Web.
• Clone it if you want the whole index offline or plan to contribute: git clone git@github.com:YsYeoh/agent-skills-hub.git
• Install a skill you find. Pick a row, open its source repo, and copy the skill's folder (the one holding SKILL.md) into your agent's skills directory — ~/.claude/skills/ for Claude Code personal skills, or .claude/skills/ in a project if it should travel with the repo. Other agents use their own equivalent folder; the format is the same.

That last step is the whole point of the hub: it gets you from "I need to do X well" to a working skill in about two minutes.

The 8 categories, and everything inside them

1. UI/UX & Frontend Design

Interface design, design systems, landing pages, redesigns, image-to-code, and the "taste" skills that fight the generic-AI look. The category that most visibly separates a good output from a template-flavoured one.

2. Motion & Animation

Web animation, React Native and Expo animation, gestures and haptics, animation audits, and the "where should I animate" lookup. The largest category in the hub, and the one with the strongest official coverage — GSAP ships its own skill set.

3. Video & Motion Video

From-HTML video generation, templated short-form marketing video, and full production pipelines. If you have ever wanted an agent to hand back a finished 20-second clip rather than a script, this is the category.

4. Coding & Language Guidance

Language-specific best practice, code review, architecture, testing, and the debugging disciplines that stop an agent from "fixing" the wrong thing. This is the category I reach for when the work is real engineering rather than scaffolding.

5. Coding Harness & Agent Orchestration

Sub-agent delegation, context and compaction, memory, verification gates, evals — everything that makes a coding agent reliable over a long session instead of impressive for five minutes. This is the quiet category that decides whether agents ship or thrash.

6. Content & Media Generation

Images for blogs and ads, brand identity, diagrams, and the social-media machinery around them. Small category, high leverage — a good hero image is the difference between a post that gets opened and one that doesn't.

7. Research & Web

Web search, scraping, and recent-events gathering. This was the last category to fill and the one I now use most — "find out what happened in the last 30 days, with sources" is a task agents get asked constantly.

8. General / Meta Skills

Skill creation, output enforcement, and the registries that index everyone else's work. Start here if you want to write skills rather than only consume them.

The pattern I keep noticing

Most of the best skills in that list share three traits:

1. They encode a decision order. "Should this animate at all, and what's it for, before which tool, before which properties." A sequence, not a checklist.
2. They say what not to do. The animation skills don't just tell you where to add motion — they reject the places that shouldn't move. Restraint is the feature.
3. They're runnable, not theoretical. A skill that says "write tests" is advice. A skill that says "here's the TDD loop, here's how to split the work to sub-agents, verify before completion" is a workflow.

If you're curating or writing skills, hold them to those three bars.

Try it

The hub is open at github.com/YsYeoh/agent-skills-hub — an index you can browse today, no clone required.

Three ways in:

• Browse a category's LINKS.md and skim the one-line summaries. Start with Coding Harness if you care about agent reliability, or Research & Web if you care about what the model can actually see.
• Install a skill by copying its folder — the one holding SKILL.md — into your agent's skills directory.
• Contribute a skill you've found or built. One row in a LINKS.md. That's it. A single link is all it takes to help the next person.

If this saved you a few tabs, it did its job.

---

What agent skills have saved you recently? Drop a comment and I'll add anything I've missed.