The best Claude Code skills, plugins and modes (2026): what each does, setup and every link (Linux, Windows, Mac)
Before you start
- 19 min read
Boxes marked Run in terminal are what you type; plain text explains why. Check what you need first

What you get in one minute#
Claude Code gets its extra powers from three kinds of extension: skills (know-how in a folder), plugins (bundles you install from a marketplace) and modes (switches that change how Claude behaves), and this page lists the best of each with the exact command and the link.
Logo: the official Claude Code logo from Anthropic's press kit (anthropic.com/press-kit)
At a glance:
- What you get: a short, checked list of official extensions, each with what it does, who it is for, how to install it and how to see that it works.
- Time needed: about 20 minutes for the install, then 2 minutes per extension.
- Difficulty: easy. You type a few commands inside Claude Code.
- You need: Claude Code and a Pro, Max, Team, Enterprise or Console account (the free Claude plan does not include Claude Code).
- Where the links are: the section Every link in one place.
| What | Value |
|---|---|
| Claude Code version checked | v2.1.311 (GitHub release of 26 October 2026) |
| Official plugin marketplace | claude-plugins-official (github.com/anthropics/claude-plugins-official, last commit 26 October 2026) |
| Official skills repository | github.com/anthropics/skills (last commit 9 October 2026) |
| Official docs | code.claude.com/docs |
How it fits together#
You switch modes, Claude loads skills from your skills folder, plugins arrive from a marketplace, and all three feed one Claude Code session that proposes edits you approve.
Before you start#
You need a terminal and a Claude account that includes Claude Code. Some plugins also need Node.js or Python for their helper programs.
| System | What to have |
|---|---|
| Linux | A terminal, curl, Git |
| Windows | Windows 10 1809+, PowerShell; Git for Windows optional (adds Bash) |
| Mac | Terminal, Git (Xcode command line tools) or Homebrew |
Checklist:
- [ ] Your own skills are backed up (see Updates and backups).
- [ ] You work inside a Git repository so every change can be undone.
- [ ] You decided to read a plugin's source before installing it (see Security and testing).
Tested on: not run by DIY Lab Hub. Checked against the official Claude Code docs, repositories and changelog in October 2026.
Step 1: Install Claude Code#
Use the native installer: it updates itself in the background. Homebrew and WinGet installs do not, so you upgrade those yourself.
Linux#
curl -fsSL https://claude.ai/install.sh | bash
Homebrew users can run brew install --cask claude-code instead.
Windows (PowerShell)#
irm https://claude.ai/install.ps1 | iex
WinGet works too: winget install Anthropic.ClaudeCode.
Mac (Terminal)#
Same installer as Linux:
curl -fsSL https://claude.ai/install.sh | bash
Or use Homebrew instead (pick one, not both): brew install --cask claude-code. That cask follows the stable channel, about a week behind; claude-code@latest gets every release.
Check it worked: open a new terminal, go to a project folder and run:
claude --version claude doctor claude
You should see a version number such as 2.1.311 (Claude Code), a health report from claude doctor, then the Claude Code prompt. Sign in in the browser when it asks. Potential issue: if claude is not found, open a fresh terminal so the new PATH loads. The full install page is in the official setup docs.
Step 2: Add skills#
A skill is a folder with a SKILL.md file. Put it in your personal skills folder and Claude can use it in every project.
Where skills live:
| Place | Folder | Used in |
|---|---|---|
| Personal | ~/.claude/skills/<name>/SKILL.md | All your projects |
| Project | .claude/skills/<name>/SKILL.md | One repository, shared through Git |
| Plugin | inside an installed plugin | Wherever the plugin is on |
A minimal skill looks like this (the format comes from the official anthropics/skills repository):
--- name: my-skill-name description: A clear description of what this skill does and when to use it --- # My Skill Name Instructions Claude follows when this skill is active.
The description is what Claude reads to decide when to use the skill, so write it like a trigger: what it does and when. Every field is optional, but description is the one that matters. Others include when_to_use, allowed-tools, model, effort, and disable-model-invocation for skills only you should start. You can also start any skill yourself by typing / and its name. Check the skills docs for the full list.
Linux#
Create a small sample skill by hand:
mkdir -p ~/.claude/skills/hello-check cat > ~/.claude/skills/hello-check/SKILL.md <<'EOF' --- name: hello-check description: Use when the user asks for a hello check. Replies with the single word "skill-loaded". --- Reply with exactly: skill-loaded EOF
Windows (PowerShell)#
New-Item -ItemType Directory -Force "$HOME\.claude\skills\hello-check" | Out-Null @' --- name: hello-check description: Use when the user asks for a hello check. Replies with the single word "skill-loaded". --- Reply with exactly: skill-loaded '@ | Set-Content -Encoding utf8 "$HOME\.claude\skills\hello-check\SKILL.md"
Mac (Terminal)#
Same commands as Linux.
Check it worked: start claude, type /skills and look for hello-check in the list. Then type "do a hello check" (or /hello-check). The answer should be skill-loaded. Claude Code watches the skills folders, so new and edited skills appear without a restart; if you created the whole skills folder while Claude Code was open, run /reload-skills.
Delete the sample when you are done: remove the hello-check folder.
Step 3: Add plugins and marketplaces#
A marketplace is a catalog of plugins. The official one is claude-plugins-official, and you install from it with one command.
Claude Code adds the official marketplace for you the first time you start an interactive session. Inside a session, /plugin install <name>@claude-plugins-official opens the plugin's details: you see what it will install and its token cost, then pick a scope (just you, everyone in this repository, or you in this repository only). To browse, run /plugin and open the Discover tab, or search claude.com/marketplace on the web. From your normal shell, claude plugin install installs straight away at user scope.
Linux#
Inside Claude Code:
/plugin install feature-dev@claude-plugins-official
Add the official skills marketplace and install its document skills:
/plugin marketplace add anthropics/skills /plugin install document-skills@anthropic-agent-skills
Windows (PowerShell)#
Inside Claude Code the /plugin commands are the same. From PowerShell (or any shell):
claude plugin marketplace add anthropics/skills claude plugin install document-skills@anthropic-agent-skills
For an owner/repo source, Claude Code uses SSH if your key works with GitHub and HTTPS otherwise. Set CLAUDE_CODE_PLUGIN_PREFER_HTTPS=1 to always use HTTPS.
Mac (Terminal)#
Same commands as Linux.
Check it worked:
claude plugin list claude plugin details feature-dev
claude plugin list shows each plugin with its version, scope and status, and claude plugin details shows how many tokens it adds to every session. Inside Claude Code, the install summary ends with Plugin is now active. or Run /reload-plugins to apply., and the plugin's skills appear when you type / as /<plugin>:<skill>.
The best skills#
Start with Anthropic's own skills. They are small, readable and maintained.
The anthropics/skills repository holds 19 skills today (checked 2026-10-27). These are the ones most people reach for:
| Skill | What it does in one line | For whom |
|---|---|---|
pdf, docx, xlsx, pptx | Create and edit PDF, Word, Excel and PowerPoint files (source-available, not open source) | Anyone who gets files, not just code |
frontend-design | Pushes UI work away from generic templated looks | Web developers |
webapp-testing | Tests a local web app with Playwright | Developers who want browser checks |
mcp-builder | Guides you to build a good MCP server | Tool builders |
skill-creator | Creates, improves and benchmarks skills | Anyone writing their own skills |
doc-coauthoring | A structured way to write specs, proposals and docs | Writers, team leads |
claude-api | Reference for the Claude API and SDK | App builders |
theme-factory, web-artifacts-builder, canvas-design, algorithmic-art | Styling and creative artifacts | Designers |
Install the two bundles with /plugin install document-skills@anthropic-agent-skills and /plugin install example-skills@anthropic-agent-skills. The repository README says these examples are for demonstration and that you should test them yourself before relying on them.
The best plugins#
The official marketplace lists hundreds of plugins now. These few earn a place on almost every machine.
All install with /plugin install <name>@claude-plugins-official.
| Plugin | What it does in one line | For whom |
|---|---|---|
feature-dev | A guided feature workflow with agents for exploring code, designing and reviewing | Anyone building features in an existing codebase |
code-review | Reviews a pull request with several agents and confidence scoring to cut false alarms | Teams on GitHub |
commit-commands | Commit, push and open a PR with slash commands | Daily Git users |
security-guidance | Warns about risky edits and reviews the diff for injection, XSS, secrets | Everyone shipping code |
claude-security | Deep vulnerability scan of your own code with every finding challenged | Owners of larger repos |
skill-creator | The same skill builder as a plugin, with evals | Skill authors |
plugin-dev | Toolkit and skills for writing your own plugins | Plugin authors |
hookify | Turns "never do X" into a hook rule from a markdown file | Anyone who wants guard rails |
claude-md-management | Audits and improves your CLAUDE.md files | Long-running projects |
claude-code-setup | Reads your repo and recommends hooks, skills and MCP servers | New projects |
context7, github, playwright | Ready-made MCP servers: library docs, GitHub, browser | Developers |
pyright-lsp, typescript-lsp | Code intelligence (language servers) | Python, TypeScript users |
Most of the official marketplace comes from partners and other authors; Anthropic maintains a smaller core such as commit-commands, code-review, feature-dev and the language servers. Anthropic also runs a community marketplace for third-party plugins that their authors submitted:
/plugin marketplace add anthropics/claude-plugins-community /plugin install <plugin-name>@claude-community
Its README says each plugin passed an automated security scan, but still tells you to trust a plugin before you install it. An old tutorial may tell you to add anthropics/claude-code: that is only the small demo marketplace (claude-code-plugins), so install the same plugins from claude-plugins-official instead to avoid two copies.
Community picks (community, review before install)#
Not made by Anthropic. Stars and dates from GitHub, October 2026.
| Project | What it does | Activity | Install |
|---|---|---|---|
| obra/superpowers | A skills framework and development method: plan, test first, review | MIT, 305k stars, pushed 23 Oct 2026 | /plugin install superpowers@claude-plugins-official |
| wshobson/agents | A large marketplace of single-purpose plugins (languages, DevOps, security) | MIT, 41k stars, pushed 26 Oct 2026 | /plugin marketplace add wshobson/agents |
| hesreallyhim/awesome-claude-code | A hand-picked list of skills, agents, status lines and tools | 56k stars, pushed 26 Oct 2026 | A list, nothing to install |
The best modes#
Modes are the cheapest upgrade: no install, just a key or a command.
Permission modes#
The permission mode decides how much Claude may do without asking. Switch with Shift+Tab, start with claude --permission-mode <name>, or set permissions.defaultMode in your user settings.
| Mode (name on screen) | What runs without asking | Use it when |
|---|---|---|
default (Manual) | Reads only; asks before edits and commands | You are new, or the code is sensitive |
acceptEdits (Edit automatically) | Reads, file edits and common file commands | You trust the plan and watch the diff |
plan (Plan) | Reads; Claude proposes a plan and changes nothing | You want to agree on the approach first |
auto (Auto) | Everything, with background safety checks that block risky actions | Long tasks. Terminal and VS Code sessions start here from Claude Code v2.1.283 when nothing else is set |
dontAsk | Only what your allow rules cover; the rest is denied | Locked-down, repeatable runs |
bypassPermissions | Everything, no checks (--dangerously-skip-permissions) | Only inside a disposable container or VM |
Shift+Tab cycles default, acceptEdits and plan; auto and bypass join the cycle only when they are enabled, and dontAsk never does. Two safety notes from the docs: a project's .claude/settings.json cannot set auto or bypassPermissions, so a repository you clone cannot switch them on for you; and claude -p (scripts) starts in Manual. To tune what auto mode allows, see Configure auto mode.
Fast mode, effort and output style#
| Setting | How to use it | What it does |
|---|---|---|
| Fast mode | /fast, Space to toggle, Enter to confirm | Opus up to 2.5x faster, same quality, higher price, paid from usage credits. Turn it on at the start of a session |
| Effort | /effort (slider) or /effort low, medium, high, xhigh, max | How hard Claude thinks. Lower is faster and cheaper; /effort auto goes back to the model's default |
| Output style | /output-style concise or /config > Output style | Built-in styles: Default, Proactive, Concise, Explanatory (adds Insight notes) and Learning (leaves small parts for you to write) |
| Skills list | /skills | Shows every skill in your session |
Mods (new in 2026)#
A mod is a plugin with JavaScript or TypeScript code that runs inside Claude Code: it can draw a pane, add a command or step into a tool call. /diff is a built-in mod (see /plugin > Installed > Built-in). A mod runs with your full permissions and is not sandboxed, so install only mods you trust and check one first with claude plugin validate <folder>. Start with the Mods overview.
Skills in the Claude.ai and desktop apps#
Skills also work in the Claude apps on every plan, Free included, as long as code execution is on.
- Go to Settings > Capabilities and turn on Code execution and file creation (called "Create and edit files" in the newer layout). On Team and Enterprise, an owner allows Skills under Organization settings > Plugins & skills.
- Zip your skill folder. The folder name must match the skill's name, and the
SKILL.mdfile must be inside. - Go to Customize > Skills, click +, then + Create skill > Upload a skill, and pick the ZIP.
- Turn the skill on. It is private to you until you share it.
Menu names can change, so check Anthropic's article Using skills in Claude if a step looks different.
Security and testing#
A skill or plugin can run code and hooks on your machine, so treat an install like running a script from the internet.
- Prefer official sources. Start with
claude-plugins-official, thenclaude-community, then your own review. - Read before you install. Look at the plugin's
hooks/,.mcp.jsonandscripts/.claude plugin validate <folder>on a clone lists its hooks and calls without running it. - Keep the mode tight. Use
defaultorplanwith a new plugin. Never use bypass outside a throwaway container. - Keep secrets out of folders where you test unknown plugins.
- Look at what is loaded.
/plugin,/skills,/hooksand/mcplist everything that is active. Disable what you do not use.
Test one plugin at a time in a test repository. If it does something unexpected, start claude --safe-mode or uninstall it:
claude plugin uninstall <plugin-name>
Troubleshooting#
| Problem | Solution |
|---|---|
/plugin isn't available in this environment. | You are in claude -p or a script. Use claude plugin install ... in your shell instead. |
Plugin "<name>" not found in any marketplace | The name is wrong or its marketplace is not added. Check /plugin > Discover, add the marketplace, use name@marketplace. |
Marketplace claude-plugins-official not found | Run /plugin marketplace add anthropics/claude-plugins-official (a new machine that never ran an interactive session does not have it yet). |
| Private marketplace fails to add | Claude Code never prompts for a password. Set up gh auth login or an SSH key first; set CLAUDE_CODE_PLUGIN_PREFER_HTTPS=1 to force HTTPS. |
Run /reload-plugins to apply. | Run /reload-plugins. If it warns about the prompt cache, /reload-plugins --force applies it anyway. |
Fast mode requires usage credits | Turn on usage credits under Settings > Usage on claude.ai (Pro, Max), or ask your admin (Team, Enterprise). |
| A skill never triggers | Rewrite the description so it says what the skill does and when to use it. Names alone do not trigger. |
| Bypass or auto missing from Shift+Tab | By design: they join the cycle only when enabled. Start with claude --permission-mode auto (bypass only in a throwaway container). |
Diagnostic commands: claude --version, claude doctor, claude plugin list, claude plugin validate <folder> (for your own plugin), and /status inside a session.
Updates and backups#
Update plugins on purpose and keep a copy of the skills you wrote.
Update commands:
claude update claude plugin update <plugin>@<marketplace> claude plugin marketplace update
The native install of Claude Code updates itself; Homebrew (brew upgrade claude-code) and WinGet (winget upgrade Anthropic.ClaudeCode) do not. Plugins from claude-plugins-official update automatically; community and third-party marketplaces do not, unless you turn auto-update on in /plugin > Marketplaces. claude plugin marketplace update with no name only refreshes the listings; to update every plugin from one marketplace, use Update marketplace on that tab. After an update you see Plugin updated: <name> · Run /reload-plugins to apply.
Backups: your own skills are plain folders, so they are easy to save. Back up only the skills folder, never your settings or login files (they can hold keys). The sample below uses the throwaway hello-check skill from Step 2.
Linux#
mkdir -p ~/backups tar -czf ~/backups/claude-skills-$(date +%F).tar.gz -C ~/.claude skills tar -tzf ~/backups/claude-skills-$(date +%F).tar.gz | head
Windows (PowerShell)#
New-Item -ItemType Directory -Force "$HOME\backups" | Out-Null Compress-Archive -Path "$HOME\.claude\skills" -DestinationPath "$HOME\backups\claude-skills-$(Get-Date -Format yyyy-MM-dd).zip" -Force
Mac (Terminal)#
Same commands as Linux.
Schedule: add the Linux command to crontab -e at 03:00 (0 3 * * * ...) on Linux or Mac, or a Task Scheduler task on Windows. Restore test: unpack the archive into an empty folder, check that skills/hello-check/SKILL.md is inside, then copy it back. Plugins need no backup: reinstall them from the marketplace.
Performance and resources#
Skills are cheap until used. Claude loads only each skill's name and description at the start (the listing cuts each one at 1,536 characters, so put the key use first). The body of SKILL.md is read when a task matches. So 50 well-described skills cost a few thousand tokens of listing, not 50 documents. Keep a SKILL.md under about 500 lines and move long reference material into extra files. For plugins, claude plugin details <name> shows the tokens a plugin adds to every turn, the /plugin panel shows the same estimate before you install, and /context shows what fills your window. The Installed tab also groups plugins you have not used recently.
MCP servers and hooks in a plugin cost more: they start helper processes, add tool descriptions to each request, and a slow hook slows every tool call. Turn off what you have not used for a month.
FAQ#
What is the difference between a skill and a plugin? A skill is one instruction folder. A plugin is a package that can contain skills, commands, agents, hooks and MCP servers, and it installs and updates through a marketplace.
Are skills only for Claude Code? No. The same SKILL.md format works in the Claude apps and the API (the open Agent Skills standard).
Is it safe to install a plugin from the official marketplace? Safer than a random repository, not risk free: most plugins there come from partners. Read the source and start in default or plan.
Which permission mode should I use every day? plan for new work, then acceptEdits or auto once you agree with the plan. Keep bypass for a disposable container.
How do I remove everything again? claude plugin uninstall <plugin> for plugins and claude plugin marketplace remove <name> for a marketplace (this also uninstalls every plugin you installed from it). Delete your own skill folders under ~/.claude/skills.
Every link in one place#
This is the dedicated list: every extension and mode from this page with its command and link.
Skills#
| Name | What it does | Install or switch | Link |
|---|---|---|---|
| Anthropic skills repository | All official example skills, the spec and a template | /plugin marketplace add anthropics/skills | github.com/anthropics/skills |
| document-skills (pdf, docx, xlsx, pptx) | Create and edit Office and PDF files | /plugin install document-skills@anthropic-agent-skills | skills/ |
| example-skills | Design, testing, MCP building, writing skills | /plugin install example-skills@anthropic-agent-skills | skills/ |
| claude-api | Claude API and SDK reference | /plugin install claude-api@anthropic-agent-skills | skills/claude-api |
| Your own skill | Any folder with a SKILL.md | Put it in ~/.claude/skills/<name>/ | Skills docs |
Plugins#
| Name | What it does | Install or switch | Link |
|---|---|---|---|
| Official marketplace | Anthropic and partner plugins | /plugin install <name>@claude-plugins-official | claude-plugins-official |
| Community marketplace | Reviewed community plugins | /plugin marketplace add anthropics/claude-plugins-community | claude-plugins-community |
| feature-dev | Guided feature workflow | /plugin install feature-dev@claude-plugins-official | plugins/feature-dev |
| code-review | Multi-agent pull request review | /plugin install code-review@claude-plugins-official | plugins/code-review |
| security-guidance | Security warnings on edits | /plugin install security-guidance@claude-plugins-official | plugins/security-guidance |
| hookify | Guard rails as markdown rules | /plugin install hookify@claude-plugins-official | plugins/hookify |
| plugin-dev | Write your own plugins | /plugin install plugin-dev@claude-plugins-official | plugins/plugin-dev |
| superpowers (community) | Skills-based dev method | /plugin install superpowers@claude-plugins-official | obra/superpowers |
| wshobson/agents (community) | Many small plugins | /plugin marketplace add wshobson/agents | wshobson/agents |
| awesome-claude-code (community) | Curated list | Nothing to install | hesreallyhim/awesome-claude-code |
Modes#
| Name | What it does | Install or switch | Link |
|---|---|---|---|
| Permission modes | default, acceptEdits, plan, auto, dontAsk, bypassPermissions | Shift+Tab, or claude --permission-mode <name> | Permission modes docs |
| Plan mode | Plan first, change nothing | Shift+Tab until "plan" shows | Permission modes docs |
| Auto mode | Background checks approve routine actions | Shift+Tab, or claude --permission-mode auto | Configure auto mode |
| Fast mode | Opus up to 2.5x faster, higher price | /fast | Fast mode docs |
| Effort | How hard Claude thinks | /effort | Model configuration |
| Output styles | Proactive, Concise, Explanatory, Learning | /output-style <name> or /config | Output styles docs |
| Mods | Code plugins that draw panes and add commands | /plugin > Installed > Built-in | Mods overview |
Docs#
| Name | What it does | Install or switch | Link |
|---|---|---|---|
| Claude Code install | Native installers for every system | curl -fsSL https://claude.ai/install.sh | bash | Setup docs |
| Install plugins | Marketplaces, scopes, updates | /plugin | Install and manage plugins |
| Anthropic's marketplaces | Official, community and demo | /plugin marketplace add <owner/repo> | Anthropic's marketplaces |
| Plugin security | What to check before installing | None | Plugin security and trust |
| Web catalog | Search all plugins, install counts | Copy the install command | claude.com/marketplace |
| Create plugins | Build and test a plugin | claude plugin validate | Create a plugin |
| Skills in the Claude apps | Settings > Capabilities | Upload a skill zip | Using skills in Claude |
| Agent Skills standard | The open format | None | agentskills.io |
| Claude Code releases | What changed in each version | claude update | GitHub releases |
Sources#
- anthropics/claude-code README and CHANGELOG (read 2026-10-27, v2.1.311)
- anthropics/claude-plugins-official README and marketplace.json (read 2026-10-27)
- anthropics/claude-plugins-community README (read 2026-10-27)
- anthropics/skills README and marketplace.json (read 2026-10-27)
- Claude Code docs, read October 2026: Advanced setup, Skills, Install and manage plugins, Anthropic's marketplaces, Permission modes, Fast mode, Model configuration, Output styles, Mods overview
- Using skills in Claude (Anthropic help center, read October 2026)
- Logo: Anthropic press kit, Claude Code logo, unchanged
- obra/superpowers, wshobson/agents, awesome-claude-code (community, read 2026-10-27)
Checked against the official docs on 2026-10-27