Claude skills and agent skills, actually reviewed.

What are Claude skills?

Claude skills are folders of instructions that Claude loads when a task calls for them. Each skill is a directory with a SKILL.md file: YAML frontmatter with a name and description, then Markdown instructions, optionally next to scripts, reference documents and templates. Claude Code reads only the description up front and loads the full skill when it looks relevant or when you call it with /skill-name, according to the Claude Code docs.

The format is called Agent Skills. Anthropic developed it and released it as an open standard, so the same folder can work in Claude Code, Codex, Cursor and other agents that support it. That is why people search for “claude skills”, “agent skills” and “anthropic agent skills” and mean the same thing. For the longer explanation, read what are Claude skills.

This site is a directory of Claude Code skills and other agent skills that someone has opened and read. It covers 195 repositories and 88 single skills. Install counts tell you what is popular; they do not tell you what a skill does to your machine.

How to pick a Claude skill

Start with the job, not the star count. A skill that does one thing well beats a pack of 300 you will never open. Four checks cover most of it.

1. Does it match your task? Browse by topic in the categories (design, SEO, code review, documents, security and others) or search the full list at /skills/.
2. How much expertise ships with it? A one-page prompt and a skill with reference files, scripts and tests cost very different amounts of context and deliver very different results. Our depth score tells them apart.
3. What can it touch? Skills can run scripts, call the network, pre-approve shell commands or install hooks. Look at the security status before the feature list.
4. Does it overlap with something you already run? Two planning methodologies in one session fight each other. Our review of addyosmani/agent-skills says outright that it overlaps with Superpowers and that you should pick one primary workflow.

If you want a shortlist, the best Claude skills page ranks only skills with a Recommended verdict.

How we review each skill

Every page carries three separate judgments.

Depth, 1 to 5. Depth measures how much expertise ships with the skill, judged from the repository files: 1 is a short prompt, 3 adds reference documents, templates or helper scripts, and 5 means references, scripts and tests or evals. Across our repositories, 14 score a 5 and 56 score a 4. A depth-1 skill can still be the right pick for a small job; the score says what you are getting, not whether it is good.

Verdict. Recommended, With caveats, or Skip, plus one sentence on why. Of the 195 repositories, 84 are Recommended and 109 are With caveats. The caveat is usually a paid dependency, heavy setup, telemetry or an unclear license.

Security status. One of three labels, computed by a fixed rule from facts we record for each file: No issues found, Review before installing and Broad access. Broad access means at least one of the following: it sends your keys to another service, uploads your data by default, loads instructions from a server at run time, uses your browser session, pre-approves broad shell commands, installs hooks without asking, pushes or merges without asking, turns off safety prompts, sends telemetry with no opt-out, or contains obfuscated content. Many reviews are partial because we could not download every file of a very large repository, and the page says so.

We also run every skill through NVIDIA’s SkillSpector scanner in static mode and use its hits as leads. The raw output is noisy: it recommended “do not install” for 193 of 286 popular skills, mostly because instructions mention commands like git push. Reviews are written by AI reviewers working from each repository’s files under a fixed rubric, and the status is calculated by code rather than chosen by the reviewer. The full rules are on the methodology page, and the security page explains each flag.

What a Claude skill can access

A skill is more than text. The Claude Code docs describe three ways a skill gets reach:

• allowed-tools in the frontmatter lets Claude use listed tools without asking while the skill runs. It does not restrict other tools, and project skills’ allowed-tools apply even in an untrusted folder, so the docs advise reviewing skills in a repository before you run Claude Code there.
• Inline !`command` blocks run a shell command before Claude sees the skill and paste the output in.
• Skills bundled in a plugin can arrive with hooks and MCP servers. The docs note that a plugin can run hooks and MCP servers, so read what it adds before you install it.

None of these is a reason to avoid skills. They are the reason a “Broad access” label exists on our pages and the reason to read a skill’s SKILL.md and scripts once before you install it.

Where to install Claude skills

For Claude Code, a personal skill goes in ~/.claude/skills/<name>/SKILL.md and a project skill in .claude/skills/<name>/SKILL.md; commit the project folder to share it with your team. Plugins bring skills through a marketplace: /plugin marketplace add owner/repo, then /plugin install name@marketplace.

For any agent, the skills CLI from Vercel Labs copies a skill into the right folder: npx skills add owner/repo -a claude-code, with -g for all projects and --skill name to pick one skill from a repository. It supports Claude Code, Codex, Cursor, OpenCode and many more, and sends anonymous usage data unless you set DISABLE_TELEMETRY=1.

Step-by-step instructions are in how to install Claude skills. Per-agent folders and commands for Claude Code, Codex, Cursor, GitHub Copilot, OpenCode, Gemini CLI, OpenClaw and Hermes Agent are on the platforms pages. If you want a bundle of skills, commands and hooks in one package, see Claude Code plugins.

Where to start

A few examples of what the ratings look like in practice, all with a Recommended verdict:

• anthropics/skills is Anthropic’s reference repository (depth 4). Check each folder’s license before you reuse code; the docx, pdf, pptx and xlsx skills are source-available, not open source.
• Matt Pocock’s skills are small and composable, and a few of them, such as tdd, have a clean security status.
• ui-ux-pro-max is a design reference with searchable style, palette and font catalogs (depth 4).
• Cloudflare’s security-audit skill is a depth-5 audit skill; budget a large token spend for a full run.

Prefer to read the rules before browsing? Start at methodology. Otherwise go to all reviewed skills.