700,000 Skills for AI: How the Plugin Market Exploded in 2026

In January 2025, the concept of “AI skills” barely existed. There were a few shared prompts in GitHub repositories, a few instruction files in project folders, and no dedicated marketplace. Sixteen months later, the skills ecosystem has surpassed700,000 optionsavailable on multiple marketplaces, moves millions of dollars and is redefining how professionals use artificial intelligence tools.

This article maps this explosion: the numbers, the marketplaces, the technical standard (SKILL.md), how skills work, why professionals pay for them instead of creating their own, and where the market is going. Whether you use Claude Code, Cursor, or any AI tool for work, this ecosystem directly affects the quality of what you produce.

1. The panorama: from zero to 700 thousand in less than 2 years

The growth of the AI ​​skills market follows a pattern that we have seen before in technology: slowly, then suddenly. The same pattern as app stores (2008-2010), WordPress plugins (2005-2008) and browser extensions (2010-2013).

The difference is speed. Apple's App Store took 2 years to reach 250,000 apps. The AI ​​skills market reached 700,000+ options in less than 18 months. Three factors explain this speed:

• Low barrier to entry:create a skill and write a Markdown file. No SDK, compiler, certificates or store approval required. Anyone with domain knowledge can publish
• Massive demand:With millions of professionals using AI tools daily, the demand for expertise is immediate. No one wants a generalist AI when they can have a specialist AI
• Network effect:the more skills there are, the more people use AI tools. The more people use it, the more skills are created. The cycle feeds back

The numbers for April 2026

Marketplace	Available skills	Business model
SkillsMP	700,000+	Open, free and paid marketplace
SkillHub	7,000+	Curation, focus on quality
Antigravity Awesome Skills	1,234+	Open-source curated list
Claude Cowork Marketplace	Thousands	Integrated into the Claude ecosystem
OpenAI Codex Plugin Marketplace	Released 27/mar/2026	Official OpenAI Marketplace

SkillsMP's 700,000+ include skills of all types and qualities -- from tested professional skills to single-line skills of questionable quality. This is natural in any open marketplace. What matters for the professional is to knowwhere to find quality skillsand how to evaluate what is worth installing.

2. Marketplaces: who sells skills today

SkillsMP: the giant of numbers

SkillsMP is the largest marketplace in absolute volume, with more than 700,000 skills catalogued. It works as an open repository where any creator can publish. The advantage is variety: if there is a skill for some obscure niche, it's probably there. The downside is the same as any open marketplace -- quality varies enormously. Skills with 5 stars and thousands of downloads coexist with abandoned skills that have never been tested.

The business model is mixed: free skills financed by advertising and skills paid with commission for the marketplace. The most successful creators on SkillsMP treat the platform as a distribution channel, not their main product.

SkillHub: curation as a differentiator

With "only" 7,000+ skills, SkillHub bets on curation. Each skill goes through a review process before being listed. This ensures a higher quality floor but limits volume. SkillHub is preferred by corporate teams that need minimal quality guarantees and support.

Antigravity Awesome Skills

Inspired by GitHub's "awesome lists" format, Antigravity is a curated list of 1,234+ skills considered exceptional by the community. It's not a marketplace -- it's more of a list of recommendations. Each skill listed has been tested and approved by community contributors. It's a great starting point for those who are starting out and want to know which skills really work.

Claude Cowork Marketplace

The marketplace integrated into the Claude ecosystem, maintained by the Anthropic community and partners. Focused on skills that work specifically with Claude Code, with native integration for installation and updates. The curation is moderate -- not as open as SkillsMP, not as tight as SkillHub.

OpenAI Codex Plugin Marketplace

The most recent and perhaps most significant: launched inMarch 27, 2026, the official OpenAI marketplace for Codex plugins. OpenAI's entry into this market validates the thesis that skills/plugins are the essential costmization layer for AI tools. We cover this in detail in section 8.

3. The SKILL.md pattern: how skills work from the inside

Every skill, in any marketplace, follows a specific technical format. The dominant pattern is theSKILL.md-- a Markdown file with YAML frontmatter that defines metadata and contains natural language instructions.

Anatomy of a SKILL.md

The YAML frontmatter

The section between---and the YAML frontmatter. It contains structured metadata that marketplaces and tools use to catalogue, search and validate the skill. The most common fields are:

• name:unique skill identifier
• description:short description for listings
• version:semantic versioning (major.minor.patch)
• author:creator or organization
• tags:search categories
• compatibility:what tools does it work with
• trigger:when the skill should be activated automatically (optional)

The body of SKILL.md

After the frontmatter comes the content itself -- instructions in natural language, formatted in Markdown. This is where the real value of the skill lies. A quality skill is not just "you are an expert in X". And a detailed document with:

• Context and role that AI should assume
• Specific evaluation or execution criteria
• Examples of expected output
• Business restrictions and rules
• Standardized response format
• Edge cases and how to deal with them

A professional SKILL.md can be 2,000 to 10,000+ words of instructions. It's not a 3-line prompt -- it's a complete specialization manual.

4. How skills work in practice

When you install a skill in Claude Code, it becomes available as a command. Installation is simple: copy the SKILL.md file to the correct folder.

The magic happens behind the scenes: when you activate the skill, the contents of SKILL.md are injected into the context of Claude Code. From that moment on, all your instructions are processed with that specialized knowledge as a basis. And how to give Claude Code an expert manual before asking him to work.

Automatic vs manual skills

Some skills are activated manually (you type the command). Others are automatic -- they activate when they detect a specific context. For example, a "code review" can be configured to activate automatically whenever you ask Claude Code to review code.triggerin frontmatter YAML defines this logic.

Skill composition

An advanced practice and combining multiple skills. You can activate a "software architecture" skill along with a "web security" and a "performance" skill. Claude Code processes all combined instructions, generating output that respects the criteria of all active skills. That's composition -- and it's what setotes basic use from professional use.

5. Skills vs prompts: the difference that defines results

The most common question from those getting to know the ecosystem: “what is the difference between a skill and a well-written prompt?” The difference andstructural, not just size.

Aspect	Prompt	Skill
Persistence	Disposable (one conversation)	Permanent (installed)
Reuse	Copy/paste every time	Activate with a command
Typical size	1-20 lines	100-5,000+ lines
Versioning	It does not have	Semantic (1.0.0, 2.1.0...)
Metadata	None	Complete frontmatter YAML
Composition	Difficult (paste multiple prompts)	Native (activate multiple skills)
Maintenance	Does not exist	Versioned updates
Depth	Superficial	Specialist (criteria, examples, edge cases)

A prompt is an instruction. A skill and ainstruction system. The most accurate analogy: a prompt and a recipe written on a napkin. A skill is a professional cooking book with techniques, times, temperatures, ingredient substitutions, reference photos and chef tips.

In practice, the difference in output is dramatic. Ask Claude Code to "audit SEO" without skill and you will receive a generic list of best practices. Request an SEO audit skill and you will receive a structured analysis with scores, priorities, correction code and updated technical references.

6. How to create your own skills

Create a basic and simple skill. Create a skillprofessionaland a research and engineering work. Here is the path for both.

Basic skill: 15 minutes

What a basic skill needs to have

• YAML frontmatter with name, description and version
• Clear definition of role/specialty
• List of criteria or rules to follow
• Expected output format

What a professional skill needs to have (and why it takes weeks)

• Deep Domain Search:It's not enough to know the basics. A "Google Ads" skill needs to cover campaign structure, bidding strategies, correspondence types, extensions, Quality Score, automation scripts, reports and dozens of other topics
• Quantitative criteria:not "use good titles", but "titles between 30-60 characters, with main keyword in the first half, expected CTR above 3.5%"
• Concrete examples:real inputs and outputs that show the expected quality
• Edge cases:What to do when the scenario is not standard
• Tests:verify that the output with the skill activated is consistently superior to the output without the skill
• Maintenance:Platforms change, algorithms update, best practices evolve. An outdated skill can be worse than no skill at all

A quality professional skill takes40-80 working hoursto research, write, test, and polish. A package of 100+ skills represents thousands of hours of accumulated work.

7. Why professionals buy ready-made skills

With 700,000 free skills available, why would anyone pay for skills? The same reason developers pay for tools instead of building everything from scratch:time and quality.

The math is simple

If you earn R$100/hour and it takes you 40 hours to research, write and test a professional skill, the cost of creating a skill is R$4,000 of your time. Buying a package with 748+ skills for $9 is 85x cheaper than creating a single equivalent skill.

But it's not just price. There are structural reasons:

• Domain Expertise:An "Advanced Meta Ads" skill written by someone who manages millions in ad spend is fundamentally different from one written by someone who read the documentation. The practical knowledge, the edge cases, the tricks that only come with experience -- that's in the skill
• Consistency:professional packages are tested as a set. The skills work well individually and in combination. Random skills from open marketplaces may conflict with each other
• Update:Platforms change constantly. A Google Ads skill that was not updated after the Performance Max change is an outdated skill. Professional packages are maintained
• Complete coverage:Instead of assembling a Frankenstein of skills from different sources, you have a cohesive package that covers an entire area of ​​expertise

8. OpenAI Codex Plugin Marketplace: The New Competitor

Em March 27, 2026, OpenAI has officially launched its Codex Plugin Marketplace -- a marketplace dedicated to plugins and skills for Codex, OpenAI's coding tool. The launch was not a surprise for those who follow the market, but the scale and ambition were greater than expected.

What changes with the entry of OpenAI

When the world's largest AI company launches a plugin marketplace, three things happen:

• Market validation:OpenAI investing significant resources in a skills/plugins marketplace confirms that this layer of costmization is not a passing trend -- and permanent infrastructure
• More creators:OpenAI's user base is massive. Thousands of new breeders will enter the market, increasing both volume and competition for quality
• Standardization:The trend is for the skills/plugins format to converge towards more uniform standards, facilitating portability between platforms

The format of the Codex Plugin

The OpenAI marketplace uses a proprietary format for Codex plugins, different from the standard SKILL.md. In practice, the content is similar -- instructions in natural language with structured metadata -- but the technical format differs. This creates a fragmentation scenario similar to what we saw with native apps (iOS vs Android): the knowledge is portable, but the packaging changes.

For professionals, this means that quality skills originally written for Claude Code can be adapted for the OpenAI ecosystem and vice versa. Specialized knowledge is not tied to a platform -- it isportable between tools.

Impact on the existing ecosystem

Existing marketplaces (SkillsMP, SkillHub, Cowork) will not disappear. On the contrary -- OpenAI's entry expands the total market. More professionals using skills means more demand for all marketplaces. What will happen is specialization: each marketplace will focus on what it does best, instead of trying to be everything for everyone.

9. Minhaskills.io in the market context

In a market with 700,000+ skills of varying quality, the question every professional asks is: "how do I setote what works from what is rubbish?" Minhaskills.io was born to answer exactly that.

Positioning: specialized curation

Minhaskills.io is not an open marketplace. It's not a list of thousands of skills where you need to mine. They arethe Cured Mega Bundle, written and maintained by experts:

• Marketing Package (748+ skills):covers technical and on-page SEO, Google Ads (Search, Performance Max, Display), Meta Ads (creation, segmentation, CAPI), Google Tag Manager, GA4, copywriting, email marketing, analytics and CRO. Each skill was written by those who practice the discipline daily, not by those who read the documentation
• Mega Bundle (748+ skills):covers React, Next.js, Node.js, Python, REST and GraphQL APIs, databases (SQL and NoSQL), DevOps (Docker, CI/CD, AWS), testing, web security, performance, accessibility and dozens of other technologies. Skills written with production standards, not tutorials

Why Mega Bundle instead of a marketplace

The decision to offer curated packages rather than an open marketplace is intentional. In a marketplace, the responsibility for choosing lies with the user. In a curated package, the responsibility for quality lies with the creator. Each skill in the minhakills.io packages:

• Has been researched with real domain depth
• Includes quantitative criteria, not just vague guidelines
• It has been tested on Claude Code to ensure consistent output
• Works individually and in combination with other skills in the package
• And updated when platforms or best practices change

Price as a filter

$9 per package is a deliberately affordable price. The goal isn't to maximize revenue per sale -- it's to maximize adoption. The more professionals use quality skills, the better the industry's collective output. And a professional who experiences the impact of quality skills in their work will never use AI again without them.

10. The future of the skills market

Looking ahead to the next 12-24 months, some trends are clear:

Consolidation of marketplaces

700,000 skills is a lot -- and many are redundant or of low quality. The market will naturally consolidate. Curated marketplaces will differentiate themselves from open marketplaces. Quality creators will migrate to platforms that value their work. Abandoned skills will be filtered or removed.

Skills as a service (skills SaaS)

Instead of purchasing single packages, subscription models for skills will emerge. You pay monthly and have access to skills that are always updated, new skills released regularly and support. This is already starting to happen in corporate niches.

Smart Skills

The next generation of skills will not be static. Smart skills will adapt their instructions based on the project context, the user's history and the type of task. Instead of a fixed SKILL.md, the skill will have conditional logic: "if the project is React, use these instructions; if it is Python, use these others."

Cross-platform standardization

With OpenAI, Anthropic and other players investing in skills/plugin ecosystems, the pressure for standardization will increase. A universal skills format that works in any AI tool would be the equivalent of HTML for the web -- a lingua franca that allows for universal creation and consumption.

Quality regulation

As skills influence professional decisions (diagnoses, legal opinions, financial recommendations), the issue of quality and responsibility will enter the regulatory radar. Who is responsible when a skill generates an incorrect recommendation that causes harm? There is no answer to this question yet, but it will need one.

The "skill-augmented" professional

The future belongs to the professional who combines domain expertise with quality AI skills. It is not the professional who knows everything, but the one who knows how to use the right tools to amplify what he knows. Skills are the link between human knowledge and AI capacity. Whoever masters this combination has a competitive advantage that will only grow.

The AI ​​skills market is not a bubble. And a new layer of infrastructure for knowledge work. Just as no one questions the existence of app stores anymore, in a few years no one will question the existence of skills marketplaces. The only question will be: are you using the right skills or not?