Zum Hauptinhalt springen

The Most Useful Skills & Plugins

Fortgeschritten
What you'll learn
  • Know the difference between a skill and a plugin, and how each gets installed
  • Add the official, community, and demo marketplaces and install plugins from them
  • Pick the right official Anthropic plugins for source control, security, LSP, and git workflows
  • Curate skills by job-to-be-done: dev, writing, research, data, ops
  • Find more from the famous community lists without getting buried in low-signal forks

The fastest way to level up Claude Code is to stop hand-writing everything and install capabilities other people already perfected. Skills give Claude on-demand expertise; plugins bundle skills, agents, commands, hooks, and MCP servers into one versioned, installable unit. This page is a curated, install-ready shortlist — what each thing does, when to reach for it, and the exact command to add it.

If you're fuzzy on the mechanics, read Skills and Plugins & Marketplaces first — this page assumes you know what a SKILL.md is.

Skill vs plugin: the 20-second model​

A skill is a folder with a SKILL.md (frontmatter description + instructions). Claude loads it only when relevant, or you invoke it with /skill-name. It lives in ~/.claude/skills/ (personal), a project's .claude/skills/, or inside a plugin. Per the official docs, custom slash commands have been merged into skills — a .claude/commands/deploy.md and a .claude/skills/deploy/SKILL.md both create /deploy.

A plugin is a distributable bundle that can ship skills, agents, hooks, MCP servers, and LSP servers together, with a .claude-plugin/plugin.json manifest. You get plugins from a marketplace — a catalog you add once, then install from. Plugin skills are namespaced, e.g. /commit-commands:commit.

Term shown.
1 / 5

Add a marketplace, install a plugin​

The official Anthropic marketplace (claude-plugins-official) is available automatically. Two others are worth adding manually: the curated community marketplace and the demo marketplace of example plugins.

Guided walkthrough1 of 4
  1. Run /plugin marketplace add anthropics/claude-plugins-community and /plugin marketplace add anthropics/claude-code. Adding a marketplace only registers the catalog — nothing is installed yet.

Bootstrap your marketplaces (run inside Claude Code)

/plugin marketplace add anthropics/claude-plugins-community
/plugin marketplace add anthropics/claude-code
/plugin
Watch out
  • Plugins and marketplaces are highly trusted components — they can execute arbitrary code with your privileges. Only add marketplaces and install plugins from sources you trust, and review the 'Will install' list first.
  • Plugin skills only appear after /reload-plugins (or a restart). If a skill still won't show, clear the cache: rm -rf ~/.claude/plugins/cache, then reinstall.

Official plugins worth installing first​

These ship in Anthropic's official marketplace (claude-plugins-official) and are the highest-signal starting point. Install with /plugin install <name>@claude-plugins-official.

Code intelligence (LSP) — make Claude see type errors instantly​

LSP plugins wire a language server into Claude Code, so after every edit Claude gets automatic diagnostics (type errors, missing imports) and precise navigation (go-to-definition, find references) instead of grep guesses. They require the matching language-server binary on your machine.

  • When to use: any non-trivial coding session in a typed language. This is probably the single biggest accuracy upgrade for dev work.
  • Trade-off: servers like rust-analyzer and pyright can use a lot of memory on big repos; disable if it bites.

Install LSP for your stack

/plugin install typescript-lsp@claude-plugins-official   # needs: typescript-language-server
/plugin install pyright-lsp@claude-plugins-official       # needs: pyright-langserver
/plugin install rust-analyzer-lsp@claude-plugins-official # needs: rust-analyzer
/plugin install gopls-lsp@claude-plugins-official         # needs: gopls

Git & PR workflow​

  • commit-commands — git workflows: stage, generate a commit message, push, open a PR. Use when you want clean, consistent commits without typing the ceremony. Skills are namespaced, e.g. /commit-commands:commit.
  • pr-review-toolkit — specialized agents for reviewing pull requests. Use before merging, or to get a second set of eyes on a diff.

Git/PR essentials

/plugin install commit-commands@claude-plugins-official
/plugin install pr-review-toolkit@claude-plugins-official
/reload-plugins

Automatic security review​

  • security-guidance — reviews each change Claude makes for common vulnerabilities and instructs Claude to fix them in the same session. Use it as a standing safety net on any code-writing project; it complements (doesn't replace) a real review.

Always-on security pass

/plugin install security-guidance@claude-plugins-official

External integrations (MCP, pre-configured)​

These bundle a ready-to-go MCP server so you skip the manual setup. Reach for one when you want Claude to act in that service directly.

  • Source control: github, gitlab
  • Project management: linear, notion, asana, atlassian (Jira/Confluence)
  • Infra: vercel, firebase, supabase
  • Design / comms / monitoring: figma, slack, sentry

Connect Claude to your tools

/plugin install github@claude-plugins-official
/plugin install linear@claude-plugins-official
/plugin install supabase@claude-plugins-official

Build-your-own toolkits & output styles​

  • plugin-dev — a toolkit for authoring your own plugins. Use the moment your .claude/ setup is worth sharing.
  • agent-sdk-dev — tools for building on the Claude Agent SDK.
  • explanatory-output-style / learning-output-style — change how Claude responds: teach-as-it-goes explanations, or an interactive learning mode. Great for onboarding or when you want to actually understand the changes.

High-value skills by job-to-be-done​

Skills shine for repeatable, context-loaded tasks. Anthropic publishes a public reference set in the anthropics/skills repo — the same patterns that power Claude's document and creative capabilities. You can copy a skill folder into ~/.claude/skills/ (personal) or .claude/skills/ (project) and it loads on the next session.

JobSkill (from anthropics/skills)When to reach for it
Documentsdocx, pdf, pptx, xlsxGenerate or edit real Office/PDF files — reports, decks, spreadsheets, fillable forms. (Source-available reference implementations.)
Build UIsweb-artifacts-builder / artifacts-builderProduce self-contained HTML/JS artifacts and mini-apps.
Extend Claudemcp-builderScaffold a new MCP server when no integration exists yet.
Brand & commsbrand-guidelines / internal-comms style skillsKeep tone, naming, and formatting on-brand across outputs.
Pro tip
  • Don't dump 30 skills into ~/.claude/skills at once. Skills are progressive-disclosure — only the name + description load until triggered — but a bloated, vaguely-described pile makes triggering unreliable. Install a few, write sharp "Use when…" descriptions, and grow deliberately.
  • AILmanac ships ready-made skill packs you can drop in to try — see /docs/templates/skills.

A self-triggering skill you can paste right now​

For research-heavy work, a skill that grounds Claude in fresh sources beats re-explaining your method every time. Drop this in ~/.claude/skills/research-brief/SKILL.md:

~/.claude/skills/research-brief/SKILL.md

---
name: research-brief
description: Use when the user asks for a researched brief, literature scan, or "what's the current state of X" with citations.
---

# Research brief

When asked for a brief on a topic:
1. Search broadly, then narrow to authoritative/primary sources.
2. Synthesize in your own words — never paste source text verbatim.
3. Flag uncertainty and date-sensitive claims explicitly.
4. End with a "Sources" list of real, clickable URLs you actually used.

Where the community lives (use the curated lists, skip the noise)​

There are dozens of "awesome" repos; most are auto-generated mirrors. A few are genuinely curated and maintained — start there.

  • hesreallyhim/awesome-claude-code — the best-known hand-curated index of skills, hooks, slash-commands, agent orchestrators, applications, and plugins. Tens of thousands of stars; actively reorganized.
  • ComposioHQ/awesome-claude-skills — a large curated catalog of Claude Skills grouped by function (documents, dev, data, marketing, security, SaaS automation).
  • VoltAgent/awesome-agent-skills — 1,400+ community + official skills (Anthropic, Vercel, Stripe, Cloudflare, Sentry, Figma, and more), portable across Claude Code, Cursor, Gemini CLI, Codex, and others.
  • anthropics/claude-plugins-community — the official community plugin marketplace; third-party plugins that passed Anthropic's automated validation and safety screening.
Watch out
  • Star counts and 'most comprehensive' claims on these repos are volatile and sometimes inflated by AI-generated forks. Treat the numbers as rough, judge by maintenance and curation, and always review a skill or plugin's actual contents before installing.
Key takeaways
  • Skill = one on-demand capability (SKILL.md); plugin = a versioned bundle (skills + agents + hooks + MCP/LSP) installed from a marketplace.
  • Add marketplaces with /plugin marketplace add <owner/repo>, install with /plugin install <name>@<marketplace>, then /reload-plugins.
  • Biggest day-one wins: an LSP plugin for your language, commit-commands, security-guidance, and the MCP integration for the service you live in.
  • Skills are portable and progressive-disclosure — curate a sharp few rather than hoarding hundreds.
  • Trust is non-optional: plugins and marketplaces run code with your privileges. Review before you install.

Check yourself

0/4
  1. What is the correct two-step flow to use a plugin from someone else's catalog?
  2. Why might a freshly installed plugin's skill not show up?
  3. What's the single biggest accuracy upgrade for typed-language dev work?
  4. Which statement about skills is accurate per the official docs?

Sources & further reading​