Where to Find Skills for Claude Code: A Map of What's Out There

The Skills Landscape

If you use Claude Code long enough, you will eventually want more than the built-in commands. Skills repositories are where that happens.

The idea is straightforward: someone has already packaged the prompts, tool configurations, and workflow patterns that make Claude Code handle a specific domain well. You import their skill, and Claude Code suddenly knows how to do something it did not know how to do before. The problem is that the skills landscape is not centralized. It is scattered across official channels, community GitHub repos, and individual developers who publish what they built for themselves. Finding what exists, and figuring out what is actually maintained, is its own work.

Built-in Skills

Anthropic maintains a set of skills that ship with Claude Code directly. Testing, architecture review, refactor — these come ready to use with a flag. They are the baseline. If you are doing something standard, start here before looking anywhere else.

claude --skill testing
claude --skill architecture-review
claude --skill refactor

The built-in skills are the most reliable because they are maintained alongside Claude Code itself. When Claude Code updates, the built-in skills update too. The downside is coverage — they only cover the most common use cases.

CCE's Published Skills

The ClaudeCodeExpert project publishes skills that cover areas the built-in ones do not address. Content workflows, specific MCP integrations, domain-specific patterns that come up in the CCE use cases. These are documented and maintained — not just throwaway examples. If you are running Claude Code in a content, marketing, or e-commerce context, the CCE skills are worth bookmarking.

The CCE skills tend to be more practical than theoretical. They were built to solve real problems rather than demonstrate a concept, which tends to mean better edge-case handling and clearer documentation.

Hermes Agent's Skill Library

Hermes Agent has one of the more extensive skill collections available. It is not Claude Code specific — Hermes Agent is a different agent framework — but the skills port over reasonably well since the underlying patterns are similar. A skill that works in Hermes Agent usually has a direct equivalent that works in Claude Code.

What Hermes has that most other repos do not is categorization and skill management tooling. The skills are organized by domain, versioned, and have clear attribution for who maintained them. This matters when you are trying to figure out whether a skill is still being maintained or whether it was abandoned two years ago.

Community GitHub Repos

The community has built a scattered collection of skill repositories across GitHub. Some are single-purpose tools someone built to solve a specific problem at their company. Others are more curated collections with multiple skills and documentation. Quality varies significantly.

The ones worth watching are the ones with active maintenance. Issues being addressed. Features being added. Documentation being updated. A repository that was updated last week tells you something a repository that has not been touched in a year does not.

The danger with community repos is that a skill nobody maintains eventually breaks in ways nobody fixes. Claude Code updates, and a skill that depended on specific tool behavior may stop working. An unmaintained skill is a liability.

Individual Developer Publications

Some developers publish skill configurations as part of their own tools, blogs, or open source projects. This is less organized — you are hunting through blog posts and gists rather than browsing a library — but the upside is specificity. A skill someone built for their own production use has usually been tested against real scenarios rather than hypothetical ones.

These are harder to find but higher quality on average. The signal is whether the developer is still using and updating the skill themselves. If they are, the skill tends to be production-grade. If they published it once and moved on, the odds of it being stale are high.

What This Means for Your Workflow

The number of skill repositories matters less than the number of skills that actually work well in your workflow. A library of fifty skills that do not do what you need is less useful than five skills that do exactly what you want.

The practical approach: start with what is built in, watch for what CCE or Hermes publish that matches your work, then look at community repos when you hit a gap those do not fill. Most people end up using three to five skills regularly and ignoring the rest.