AgentSkills.codes

What AgentSkills.codes is

AgentSkills.codes is a free, open registry of Agent Skills: the markdown instruction files that teach an AI coding assistant a workflow or a body of domain knowledge. It indexes 19,296 installable skills across nine categories, the largest being Coding with 6,343 of them, followed by Productivity and Planning. Everything in it is scraped from public GitHub repositories rather than submitted by vendors, and the site says it re-scans the entire catalog every day for freshness and safety signals.

The important thing to understand about the format is that a skill is not a plugin or an API integration. It is a markdown file with YAML frontmatter that gets copied into your agent's skills directory, which is why the same file works across Claude Code, Codex, GitHub Copilot and Antigravity. That portability is the whole point: you can fork a skill, edit it, or vendor it into your own repo and stop depending on the registry entirely.

Pricing

Free, and structurally so. The registry is open source under the MIT license. There are no paid tiers, no accounts required to browse, no usage limits, and no upgrade path, because there is no commercial product behind it. The skills it lists are public repositories with their own licenses, which the registry surfaces on each listing.

This matters for a practical reason rather than a philosophical one: there is no vendor with an incentive to keep listings alive, promote particular entries, or lock you in. It also means nobody is contractually on the hook if a skill in the index turns out to be junk.

Where it is the strongest pick

Breadth and provenance. If you already know roughly what you want the agent to do and want to see whether somebody has published something for it, an index this size answers that faster than searching GitHub by hand. The per-listing metadata is the genuinely useful part: source repo, star count, license and last commit date, all on the card, so you can dismiss an abandoned skill without opening it.

It also does not pretend to be a curated best-of. That is a reasonable design choice for a scraped catalog, and it is more honest than a directory that implies editorial review it never performed.

Skills are not MCP servers

This trips people up constantly, and it changes which registry you should be searching. A skill is a markdown file: instructions, conventions and examples that shape how the agent works. It runs nothing, needs no process, and costs nothing but context. An MCP server is software you connect to the agent so it can call out to a live system, such as a database, a ticketing tool or an API.

If you want the agent to write commit messages your team's way, that is a skill. If you want it to read your actual Jira tickets, that is an MCP server. Registries for the two are separate, and searching the wrong one is the most common reason people conclude "there is nothing for this." AgentSkills.codes is squarely a skills index.

Best for

Developers already using Claude Code, Codex or Copilot who want a searchable index of community skills, and anyone comparing what exists before writing a skill themselves. It is a discovery layer, not a curation layer.

Key features

• 19,296 indexed skills across nine categories, Coding being the largest at 6,343
• Automatic daily re-scan of the full catalog for freshness and safety signals
• Per-skill provenance: GitHub source link, stars, license and last commit date
• One-command install that copies the markdown into your agent's skills directory
• Cross-agent compatibility, including Claude Code, Codex, GitHub Copilot and Antigravity
• MIT licensed and open source, with no accounts or paid tiers

Best-fit use cases

• Finding an existing skill for a narrow task before spending time authoring one
• Auditing what the community has converged on for a workflow you are formalizing
• Sourcing a starting point to fork and adapt into your own repo
• Checking whether a skill you already installed is still maintained upstream

Pros

• Genuinely free with no commercial path, so no upsell pressure or listing bias
• Provenance on every listing makes abandoned skills easy to spot
• Daily re-scanning keeps stars, licenses and commit dates from going stale
• Skills are portable markdown, so you are never locked into the registry
• Cross-agent, not Claude-only

Cons

• Scraped, not curated: inclusion signals existence, not quality
• Nearly 20,000 entries means the hard problem shifts from finding skills to judging them
• No editorial ranking, reviews or usage data to separate the good from the abandoned
• No named operator on the site, so there is no accountable maintainer if it goes dark
• Nothing is portable-by-default about quality: a bad skill installs exactly as easily as a good one

What to watch out for

A skill is instructions your agent will follow. That is a different security posture from a read-only directory listing. Before installing anything from any registry, including this one, read the markdown, and be especially careful with skills that touch credentials, delete files, or run shell commands. Automated safety scanning reduces the obvious risks; it does not replace reading a file that is, by design, short enough to read.

Also worth knowing: because the catalog is scraped from GitHub, the same skill can appear multiple times under different forks. Check the star count and commit date to find the upstream original rather than someone's stale copy.

Is AgentSkills.codes worth using?

As a free discovery tool, yes, with a clear-eyed view of what it is. It solves "does a skill exist for this" well and "which skill should I trust" not at all, and it is upfront about that. Most developers will end up using it alongside Anthropic's own skills repository, which is small and first-party, and a curated list or two for the judgment call. Our roundup of Claude Code skill directories covers how the options differ.

There is no paid tier to weigh up and nothing to cancel, so the only real cost of trying it is the time to evaluate whatever you install.

FAQ

Is AgentSkills.codes free?

Yes. The registry is free and open source under the MIT license, with no paid tiers, no accounts required to browse, and no upsell path. The skills it indexes are themselves public GitHub repositories, so you can fork, edit or vendor any of them into your own repo.

Where do the skills come from?

They are scraped from public GitHub repositories rather than submitted by vendors. Each listing links back to its source repo and shows stars, license and last commit date, and the registry says it re-scans the entire catalog every day for freshness and safety signals.

Does it only work with Claude Code?

No. Agent Skills are markdown instruction files, so they are portable. The registry lists compatibility with Claude Code, Codex, GitHub Copilot and Antigravity among others. Installation copies the markdown into your agent's skills directory.

Is a scraped registry safe to install from?

Treat it like any package registry: the catalog is automatically collected rather than hand-reviewed, so inclusion is not an endorsement. The registry runs safety scanning and surfaces provenance signals, but a skill is instructions your agent will follow, so read the markdown before installing anything that touches credentials, deletes files or runs shell commands.

How is it different from Anthropic's own skills repo?

Anthropic's repository is a small, first-party set of reference skills that demonstrate the format. AgentSkills.codes is a broad third-party index of what the wider community has published. Most people use the official repo to learn the pattern and a registry like this one to find something specific.

Verified August 2026 from the vendor's own site. AgentSkills.codes has no affiliate program and no commercial relationship with ToolChase; the link above is a plain, untracked link.