Agent Skills marketplaces, directories and registries

As Agent Skills spread across ecosystems, a natural question follows: where do you find, share and trust them? This guide explains the difference between directories, marketplaces and registries — what each is good for, what to expect from one, and what could emerge next.

Note: AISkills.guru does not operate a skills marketplace. This is an educational overview of the category.

For the format itself, see What are AI Agent Skills? and the SKILL.md guide.

Directory vs marketplace vs registry

The terms are often used loosely, but they point at different things:

• A directory is an index — a place to discover skills, usually with metadata and links. It answers "what exists?".
• A marketplace adds transaction: discovery plus install, licensing, and sometimes paid distribution. It answers "what exists, and can I get it here?".
• A registry is infrastructure — a versioned, resolvable source of packages, closer to a package manager backend. It answers "give me version X of this skill, reliably".

A single platform can be more than one of these. The distinctions matter because they imply different responsibilities around trust and safety.

What users might expect

As the category matures, people look for several things from any central place for skills:

• Discovery — search and browse by task, ecosystem or keyword.
• Installation — a simple path from "found" to "loaded".
• Versioning — knowing which version you have, and updating safely.
• Security — signals about what a skill can do before you install it.
• Reviews and reputation — community signal on quality and safety.
• Provenance — who authored the skill, and is it the real one?
• Compatibility — which agents and versions a skill supports.

What makes a useful registry

Trust is the hard part. Because skills can contain scripts that run on your machine, a good registry earns trust through:

• Clear provenance — verified authors and packages.
• Versioning and change history so updates are reviewable.
• Security posture — disclosure of scripts, network calls and permissions.
• Compatibility metadata so you know what a skill targets.
• A path for reporting and removing malicious skills.

These are the same lessons package ecosystems learned the hard way. Skills are code-adjacent, and the security guide covers how to evaluate an individual skill.

The current state

Today, skills are mostly shared as folders — in repositories, gists, and internal collections — and installed by pointing an agent at a directory. The open Agent Skills site tracks which agents support the format. Centralised directories, marketplaces and registries are emerging but not yet settled, so there is no single canonical store.

What could emerge next

Plausible directions include: namespace and naming conventions to avoid collisions; signed skills and verified authors; automated security scanning on upload; compatibility manifests; and registries that integrate directly with agents for one-command install. None of this is guaranteed — it depends on what the ecosystem adopts.

A practical stance today

Until registries mature, treat skills the way you treat any third-party code: read the SKILL.md, inspect scripts, check provenance, and pin versions. Prefer skills you can read over opaque packages. If you want to author and share skills, the creation guide is the place to start.

Frequently asked questions

Is there an official Agent Skills marketplace? There is an open format and a site tracking support, but no single official store. Confirm current options against agentskills.io.

Can I sell skills? Whether a skill can be sold depends on the platform distributing it. The format itself is open and free to use.

How do I know a skill is safe? Inspect it before installing — read the file and any scripts, and check provenance. See Agent Skill security.

Do skills need a registry to work? No. A skill is just a folder; it works locally without any central service.