I Stopped Configuring AI Tools. Skills Do It For Me

Three weeks ago, I spent two hours teaching Cursor how we handle database migrations. Last week, I taught Claude Code the same thing — separate two hours gone.

Then I discovered the Antigravity skills library. One install later, both tools knew our migration patterns, our testing conventions, and our error handling style. No manual configuration. No copying docs between projects.

That is when it hit me: skills are becoming the new plugins.

The Numbers Are Going Crazy

Let me paint you a picture:

• Antigravity Awesome Skills: 26,218 stars in 9 weeks (created January 14, 2026)
• 1,273+ community skills available right now
• Cross-platform support: Claude Code, Cursor, Codex CLI, Gemini CLI

For context, VS Code took years to build its plugin ecosystem. These skills did it in weeks.

We spent a decade building plugin systems for IDEs. AI coding assistants did it in months — with prompts instead of API bindings.

— ***

Why Skills Are Different This Time

Here is what gets me: these are not plugins in the traditional sense. There is no SDK, no API to learn, no type definitions to import.

A skill is literally just:

1. A prompt that tells the AI how to act
2. Some reference files it loads when activated

That is it. And because they are prompts, they:

• Work everywhere — no platform-specific bindings
• Are easy to write — if you can write a GitHub README, you can write a skill
• Adapt to context — the same skill behaves differently based on what it detects

Here is what I mean:

You: "Build a new API endpoint"
-> Skill detects: new endpoint being created
-> Loads: references/api-conventions.md
-> Applies: your middleware patterns, error handling, validation rules

Versus the old way:

You: Configure your IDE
-> Install extension 1
-> Install extension 2  
-> Copy-paste config files
-> Restart IDE
-> Hope nothing breaks

See the difference? Skills are context-aware out of the box.

The Ecosystem Is More Than Just One Repo

Antigravity gets the attention, but it is not alone:

Repo	Stars	Focus
Antigravity Awesome Skills	26K+	Cross-platform skills library
CLI-Anything	20K	Making all software agent-native
OpenViking	17K	Context databases for agents
Planning with Files	16K	Manus-style persistent planning

This is not a niche anymore. This is an arms race to build the best skill library.

What Can You Actually Do With Skills?

Let me give you a few examples from the library:

• Full-Stack React Developer — knows your component patterns, testing setup, and deployment workflow
• Security Auditor — scans your code for OWASP vulnerabilities as you type
• Database Schema Designer — understands your ORM conventions and migration patterns
• API Documentation Generator — builds docs from your code in your team’s style
• Test Coverage Booster — finds untested code and writes tests in your testing framework

And the beautiful part? Most of these are one-line installs.

When Skills Go Wrong

I have to be honest with you — not all skills are created equal. Some issues I have run into:

1. Too generic — a skill that claims to “improve code quality” usually does not help
2. Scope creep — some skills try to do everything and end up doing nothing well
3. Outdated — skills written for older Claude versions sometimes behave oddly

My rule: if a skill does not have clear, specific triggers, skip it. The best skills I use have one job and do it extremely well.

The Future: Every Team Will Have Their Own Skill Library

I predict that within a year:

• Agencies will ship with client-specific skills pre-installed
• Open source projects will include skills in their repositories
• Enterprises will have private skill registries for internal patterns

You will be able to say “we use the internal-api skill” and every AI tool will automatically follow your conventions.

It is like having a senior developer whisper in the AI’s ear: “This is how we do things here.”

Where to Start

If you want to try skills:

1. Install the Antigravity CLI (they have an installer)
2. Start with 2-3 skills that match your daily workflow
3. Iterate — add skills for your specific pain points

Do not try to install everything at once. Pick the one thing you explain most often to AI tools and find a skill for that.

The plugin era for AI coding assistants is just beginning. You can either ride the wave or keep manually configuring everything.

I know which one I choose.

---

What skills are you most excited about? Or is there a gap in the current ecosystem that needs filling?