On this page
- What is a Claude Code skill?
- Where skills live
- What is a Claude Code plugin?
- The best Claude Code skills and plugins
- Workflow and planning
- Superpowers
- Matt Pocock's skills
- feature-dev
- Code review and Git
- code-review and pr-review-toolkit
- commit-commands
- security-guidance
- Frontend and design
- frontend-design
- Playwright
- Docs and documents
- Context7
- Anthropic document skills
- Build your own
- skill-creator
- claude-md-management
- Finding more
- skills CLI and skills.sh
- Curated lists
- Popular ones we did not keep
- How to write your own skill
Out of the box, Claude Code is a very good general coding agent. Skills and plugins are how you make it good at your particular way of working: how you plan, how you review, how you write docs, what your design system looks like.
We use Claude Code every day on this site and on about 30 other projects. This list is the set of skills and plugins we have actually kept installed, plus the popular ones we tried and why some did not stay. Every install line, repo and star count was checked on September 26, 2026.
- A skill is a folder with a SKILL.md file. Claude only reads its short description until a task needs it, so skills are cheap to keep installed.
- A plugin is a bundle: skills, slash commands, subagents, hooks and MCP servers in one install.
- Start with the official marketplace. It is added automatically: type /plugin and browse.
- Best first installs: Superpowers or Matt Pocock's skills for workflow, frontend-design for UI, skill-creator to write your own.
- Plugins can run code on your machine. Read hooks and MCP config before installing anything from a stranger.
What is a Claude Code skill?
A skill is a folder containing a SKILL.md file: a short YAML header with a name and description, followed by plain Markdown instructions. It can also carry scripts, reference docs and templates.
The clever part is how it loads. Claude keeps only each skill's description in context, about 100 tokens per skill. When your request matches a description, Claude reads the full file. When the skill needs a script or a reference doc, that loads last. So you can install 20 skills without paying for 20 full instruction files on every message. You can also run any skill directly by typing /skill-name.
Here is what one looks like. This is the header of grill-me, a skill we use before every big change:
---
name: grill-me
description: Interview the user relentlessly about a plan or design until
reaching shared understanding. Use when user wants to stress-test a plan.
---Skills are now an open standard (published at agentskills.io), so the same folder works in other agents too. OpenAI Codex, Gemini CLI, Cursor, GitHub Copilot and opencode all read SKILL.md files. The shared path .agents/skills/ is becoming the place to put skills you want every tool to see.
Where skills live
| Scope | Folder | Who sees it |
|---|---|---|
| Personal | ~/.claude/skills/<name>/ | You, in every project |
| Project | .claude/skills/<name>/ | Everyone who clones the repo |
| Plugin | inside the plugin, run as /plugin:skill | Anyone with the plugin enabled |
| Cross-tool | ~/.agents/skills/ or .agents/skills/ | Codex, Gemini CLI, Cursor, Copilot (and Claude Code through the skills CLI) |
On our machine the personal folder holds two skills we wrote ourselves, and ~/.agents/skills/ holds Matt Pocock's set, installed once and shared across Claude Code and Codex.
What is a Claude Code plugin?
A plugin is a package that can bundle several things at once: skills, subagents, hooks (scripts that run on events like "before a file edit"), MCP servers and even language servers. One install, one toggle to switch it all off.
The commands you need:
/plugin # browse, install, manage
/plugin install frontend-design@claude-plugins-official
/plugin marketplace add obra/superpowers-marketplace
/plugin disable superpowers # keep it, stop loading it
claude plugin list # same thing from your shellA marketplace is just a Git repo with a catalogue file. Anthropic's official one, claude-plugins-official, is added the first time you run Claude Code and lists over 300 plugins, most of them from partners such as GitHub, Figma, Sentry and Supabase.
- Anthropic's own docs say a plugin can run arbitrary code with your user permissions. Hooks and MCP servers run outside Claude's sandbox.
- Anthropic does not review third-party plugins, and auto-update can change files after you checked them.
- Snyk scanned 3,984 public skills in February 2026 and found 76 confirmed malicious ones, most using prompt injection.
- Before installing from an unknown repo, open
hooks/hooks.json,.mcp.jsonand anybin/orscripts/folder, or runclaude plugin details <name>.
The best Claude Code skills and plugins
Ordered by how useful they have been to us, not by stars. Stars are rounded GitHub counts from September 26, 2026.
Workflow and planning
Superpowers
The most popular skills pack on GitHub (about 292k stars), written by Jesse Vincent. It turns Claude into a methodical engineer: brainstorm first, write a plan, execute it in small steps with tests, debug systematically, and use Git worktrees and subagents for parallel work.
Install with /plugin install superpowers@claude-plugins-official. The first few sessions feel slower because Claude asks more questions. That is the point. Our experience: fewer "it compiled but did the wrong thing" moments on large changes.
- Replaces
- Ad hoc prompting for plans and TDD
- Runs on
- Claude Code, Codex, Cursor, Gemini CLI, opencode
- Difficulty
- Easy
- Licence
- MIT (open)
- Skip if
- You want Claude to just start typing code
Matt Pocock's skills
About 270k stars and the set we use most. The highlights are grill-me (Claude interviews you about a plan until every branch is decided), diagnose (a disciplined debugging loop), tdd, to-prd and to-issues (turn a conversation into a spec or GitHub issues), and zoom-out for a map of unfamiliar code.
Install with npx skills@latest add mattpocock/skills (works across agents) or /plugin install mattpocock-skills from the official marketplace, then run /setup-matt-pocock-skills once per repo. Pick this or Superpowers, not both: their planning skills overlap.
- Replaces
- Long custom planning prompts
- Runs on
- Any agent that reads SKILL.md
- Difficulty
- Easy
- Licence
- MIT (open)
- Skip if
- You already run Superpowers and like it
feature-dev
Anthropic's own guided workflow for building a feature: explore the codebase with subagents, ask clarifying questions, propose an architecture, then implement and review. A lighter, official alternative to the two packs above. /plugin install feature-dev@claude-plugins-official.
- Replaces
- One-shot feature prompts
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- You already use a workflow pack above
Code review and Git
code-review and pr-review-toolkit
code-review runs several specialised review agents over a pull request and only reports findings above a confidence score, which cuts the noise. pr-review-toolkit adds focused agents for tests, error handling, types and comments. Install both from @claude-plugins-official and run them before you open a PR, not after.
- Replaces
- A second pair of eyes you do not have
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- Your team already runs an AI reviewer in CI
commit-commands
Adds /commit-commands:commit, plus commands to push and open a pull request. Small, but it is the plugin you will run most often. If you have house rules (we require one-line commit messages with nothing else), put them in your CLAUDE.md and the command follows them.
- Replaces
- Typing commit messages
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- You have strict commit rules in CLAUDE.md already
security-guidance
Warns while Claude edits and reviews diffs for injection, XSS, SSRF, leaked secrets and more than 25 other issue types. It uses hooks, so it catches problems as they are written rather than at review time. Cheap insurance for anything that touches user input.
- Replaces
- Hoping you spot the XSS
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- You run a dedicated SAST tool on every diff
Frontend and design
frontend-design
Makes Claude commit to a real visual direction (type, colour, layout, motion) instead of the same safe template every time. It has about 924k installs on skills.sh, the top design skill there. /plugin install frontend-design@claude-plugins-official. Pair it with a design-system section in your CLAUDE.md so it respects your tokens.
- Replaces
- Generic purple-gradient AI layouts
- Runs on
- Claude Code, any SKILL.md agent
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- You only write backend code
Playwright
Microsoft's Playwright MCP server wrapped as a plugin. Claude can open your app in a real browser, click, fill forms and read the page, which is how it checks its own UI work. /plugin install playwright@claude-plugins-official. More browser options in our list of the best MCP servers for Claude.
- Replaces
- Manually clicking through your app
- Runs on
- Claude Code, Node.js
- Difficulty
- Easy
- Licence
- Apache-2.0 (open)
- Skip if
- Your project has no web UI
Docs and documents
Context7
Fetches current, version-specific documentation for thousands of libraries, so Claude stops suggesting deprecated APIs. It solves the single most common complaint we hear about AI coding on fast-moving frameworks like Astro, Next.js and Tailwind. /plugin install context7@claude-plugins-official.
- Replaces
- Claude guessing at last year's API
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- MIT (open)
- Skip if
- You work on one stable, well-known stack
Anthropic document skills
The same skills that let claude.ai create Word, PowerPoint, Excel and PDF files, published in Anthropic's anthropics/skills repo (about 178k stars). Useful for reports, invoices, data exports and filling PDF forms from the terminal.
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills- Replaces
- Copy-pasting into Word and Excel
- Runs on
- Claude Code, Python
- Difficulty
- Easy
- Licence
- Source available
- Skip if
- You never touch Office files
Build your own
skill-creator
The most valuable skill on this list long term. It interviews you, writes the SKILL.md, and can test it against example prompts to check Claude actually triggers it. We use a custom skill for writing and publishing articles on this site, and it is the one we would least like to lose.
- Replaces
- Writing SKILL.md by trial and error
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- Never
claude-md-management
Audits and tidies your CLAUDE.md files: finds contradictions, stale rules and things that belong in a skill instead. Anthropic recommends keeping each CLAUDE.md under 200 lines; ours had drifted well past that before we started pruning. See our AGENTS.md and CLAUDE.md examples for what to keep.
- Replaces
- A CLAUDE.md that only ever grows
- Runs on
- Claude Code
- Difficulty
- Easy
- Licence
- Anthropic
- Skip if
- Your CLAUDE.md is under 100 lines
Finding more
skills CLI and skills.sh
Vercel's cross-agent installer. npx skills add owner/repo installs a skill into Claude Code, Codex, Cursor, opencode and dozens more, and npx skills update keeps them current. skills.sh is the matching directory with install counts. Its most-installed skill, find-skills (about 3.6M installs), lets your agent search the directory for you.
- Replaces
- Copying skill folders by hand
- Runs on
- Node.js
- Difficulty
- Easy
- Licence
- MIT (open)
- Skip if
- You only use Claude Code plugins
Curated lists
When you want something niche, start with hesreallyhim/awesome-claude-code (about 55k stars) and ComposioHQ/awesome-claude-skills (about 76k). Trail of Bits publishes a well-regarded set of security audit skills at trailofbits/skills. Apply the same caution as any third-party code.
- Replaces
- Random skills from social media
- Runs on
- A browser
- Difficulty
- Easy
- Licence
- Various
- Skip if
- The official marketplace covers you
Popular ones we did not keep
- Big "everything" packs. Marketplaces with hundreds of agents look impressive, but every enabled skill adds its description to every message. We saw context use climb with no gain. Install the one skill you need, not the whole shelf.
- Humanizer-style skills. They rewrite text to hide AI tells. We would rather write plainly in the first place, with rules in CLAUDE.md.
- Running two workflow packs at once. Superpowers and Matt Pocock's skills both want to own planning. Claude ends up following half of each.
How to write your own skill
If you keep pasting the same instructions into Claude, that is a skill. The minimum:
- Create
~/.claude/skills/my-skill/SKILL.md(or.claude/skills/in a repo to share it). - Add a header with
name(lowercase, hyphens, same as the folder) and adescriptionthat says what it does and when to use it. The description is what triggers it, so be specific. - Write the instructions in plain Markdown. Keep the main file under 500 lines and move long references into separate files the skill links to.
- Test it by asking for the task in normal words, then with
/my-skill. If it does not trigger on its own, rewrite the description.
- Descriptions say when to use the skill, not just what it is
- No secrets or API keys inside skill files
- Scripts the skill runs are checked into Git and reviewed
- Rarely used skills are disabled, not deleted, to save context
For the other half of the setup, see our best MCP servers for Claude, and if cost is the worry, how to use Claude Code for free with OpenRouter. Comparing agents first? Read Claude Code vs Codex vs Gemini CLI vs opencode.
What is the difference between a Claude Code skill and a plugin?
Where do I put Claude Code skills?
Do skills slow Claude Code down or use a lot of tokens?
Are Claude Code skills safe to install?
Do Claude Code skills work in Codex or Gemini CLI?
Sumit
ContributorHi, I'm Sumit, Being an introvert I have always been obsessed with technology-computers and reading dozens of posts to learn, find answers out of my curiosity. I love to write as I explore more.
