Best Claude Code skills: 12 picks for shipping code

The best Claude Code skills depend on the job: Superpowers for planning and test-driven work, frontend-design for UI, Vercel’s react-best-practices for performance, and cubic’s run-review for code review. The 12 picks below include individual skills, skill bundles and built-in commands, grouped by the work they help you finish.

Agent Skills package instructions in a SKILL.md file, sometimes with scripts and references. Our guide to agent skills and how SKILL.md works explains the open Agent Skills format and how it differs from CLAUDE.md. Install commands and capabilities below were checked against publisher documentation on October 15, 2026; this is an editorial shortlist, not a benchmark of every skill.

Best Claude Code skills by task: 12 picks

Commands that name a marketplace other than claude-plugins-official need that marketplace added first, with /plugin marketplace add <owner/repo>.

Planning skills

1. Superpowers (obra/superpowers)

Superpowers is a development workflow pack for coding agents. Its skills trigger on their own, one stage after another. brainstorming asks what you want before any code is written, writing-plans breaks the approved design into tasks of 2 to 5 minutes with exact file paths and verification steps, and subagent-driven-development or executing-plans works through them. The same pack includes test-driven-development (red, green, refactor), systematic-debugging (a four-phase root cause process), requesting-code-review and using-git-worktrees.

Install it from Anthropic's official marketplace:

/plugin install superpowers@claude-plugins-official

/plugin install superpowers@claude-plugins-official

/plugin install superpowers@claude-plugins-official

It fits a feature that takes more than one sitting, where you want a written plan before any code. The README says the workflows are "mandatory workflows, not suggestions," so expect Claude to ask questions and write tests before it writes the feature. It is MIT licensed, and its README documents an optional telemetry ping you can turn off with SUPERPOWERS_DISABLE_TELEMETRY.

2. Agent Skills by Addy Osmani (addyosmani/agent-skills)

Addy Osmani’s Agent Skills pack maps skills to stages of development, with slash commands such as /spec, /plan, /build, /test and /ship. spec-driven-development writes a PRD covering objectives, structure, code style, testing and boundaries before any code. planning-and-task-breakdown turns that spec into small tasks with acceptance criteria, and interview-me asks one question at a time when a request is underspecified.

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

Use it if you want a spec on disk before Claude starts. Its review command collides with a built-in Claude Code command of the same name, so the README says to call it as /agent-skills:review.

Testing and debugging skills

3. /run and /verify (bundled with Claude Code)

Claude Code ships these with no install. Per the Claude Code skills docs, /run launches and drives your app to see a change working, and /verify builds and runs the app to confirm a change does what it should, "without falling back to tests or type checks." A third bundled skill, /run-skill-generator, records how to build and launch your project as a skill in .claude/skills/run-<name>/, so later runs follow the same recipe.

Use /verify after a change that tests don't fully cover, such as a UI flow or a CLI's output. It runs only when you invoke it, since it spends time and tokens.

4. webapp-testing (anthropics/skills)

The SKILL.md describes it as a "toolkit for interacting with and testing local web applications using Playwright." Claude writes Python Playwright scripts, and a helper script, with_server.py, manages the server lifecycle. It can take screenshots and read browser logs.

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

Ask for it when you want Claude to click through a local app and report what broke, without setting up a test framework first.

5. Playwright CLI skill (microsoft/playwright-cli)

The Playwright CLI gives coding agents browser automation commands through a skill. Its README says CLI workflows are more token-efficient than MCP because they avoid loading large tool schemas and accessibility trees into context. The installed skill includes guides for running and debugging Playwright tests, request mocking, tracing, and generating tests from a spec.

npm install -g @playwright/cli@latest
playwright-cli install --skills

npm install -g @playwright/cli@latest
playwright-cli install --skills

npm install -g @playwright/cli@latest
playwright-cli install --skills

It suits a project that already has a Playwright suite, or one where you want Claude to start one.

Frontend skills

6. frontend-design (anthropics/skills)

Its description: "Guidance for distinctive, intentional visual design when building new UI or reshaping an existing one." The instructions push Claude to pick typography, palette and layout for the specific product and to avoid the default treatments that make generated pages look alike. It ships in the same example-skills plugin as webapp-testing.

Name it in your prompt when Claude is building or redesigning a page and you want something other than the template look.

7. react-best-practices (vercel-labs/agent-skills)

Vercel’s skill collection includes React and Next.js performance rules from Vercel Engineering, "40+ rules across 8 categories, prioritized by impact," starting with eliminating request waterfalls and bundle size. The same repo has web-design-guidelines, which audits UI code against 100+ rules on accessibility, performance and UX, and composition-patterns for component APIs.

npx skills add vercel-labs/agent-skills

npx skills add vercel-labs/agent-skills

npx skills add vercel-labs/agent-skills

Use it while writing components or data fetching in a React or Next.js app, or ask Claude to review a component against it.

Database and backend skills

8. supabase-postgres-best-practices (supabase/agent-skills)

Supabase’s collection includes two skills. supabase covers its products (Database, Auth, Edge Functions, Storage and more), and supabase-postgres-best-practices holds Postgres guidance across 8 categories: query performance, connection management, schema design, row-level security and others.

npx skills add supabase/agent-skills --skill supabase-postgres-best-practices

npx skills add supabase/agent-skills --skill supabase-postgres-best-practices

npx skills add supabase/agent-skills --skill supabase-postgres-best-practices

Or as a Claude Code plugin:

claude plugin marketplace add supabase/agent-skills
claude plugin install postgres-best-practices@supabase-agent-skills

claude plugin marketplace add supabase/agent-skills
claude plugin install postgres-best-practices@supabase-agent-skills

claude plugin marketplace add supabase/agent-skills
claude plugin install postgres-best-practices@supabase-agent-skills

Use the Postgres skill on any Postgres project, Supabase or not, when Claude writes queries, indexes or migrations.

Code review skills

9. /code-review and /simplify (bundled with Claude Code)

The commands reference lists both as bundled skills. /code-review reviews the current diff, or a PR number, branch or path, for correctness bugs, with --fix to apply findings. /simplify runs four review agents in parallel on reuse, simplification, efficiency and abstraction level, then applies the fixes; the docs say it doesn't look for correctness bugs.

Use /simplify when you want cleanup and are ready for it to edit files. Use /code-review to inspect correctness bugs; add --fix only when you want changes applied. A local review is one pass. Follow it with AI code review on the pull request so follow-up commits are reviewed too.

10. cubic skills (mrge-io/skills)

Disclosure: cubic is our AI code review product, and these skills connect Claude Code to it. run-review reviews uncommitted changes or the current branch through the CLI, check-pr-comments works through unresolved PR review threads and resolves the ones it handled, and cubic-loop repeats review and fixes up to an iteration limit (five by default). Local reviews need the cubic CLI installed and signed in; PR-comment workflows need the GitHub CLI; codebase-context workflows need MCP. The cubic skills installation guide lists the prerequisites for each workflow.

/plugin marketplace add mrge-io/skills
/plugin install cubic

/plugin marketplace add mrge-io/skills
/plugin install cubic

/plugin marketplace add mrge-io/skills
/plugin install cubic

For other agents, the docs give npx @cubic-plugin/cubic-plugin install --skills-only. Our code review skill guide compares this setup with a review skill you write yourself.

Document skills

11. docx, pdf, pptx and xlsx (anthropics/skills)

These are the skills behind Claude's own file creation, published as source-available reference code rather than open source. They read and write Word, PDF, PowerPoint and Excel files.

/plugin install document-skills@anthropic-agent-skills

/plugin install document-skills@anthropic-agent-skills

/plugin install document-skills@anthropic-agent-skills

Use them when a coding task ends in a Word, PDF, slide or spreadsheet file. The README's example prompt: "Use the PDF skill to extract the form fields from path/to/some-file.pdf."

Skills for building skills

12. skill-creator and mcp-builder (anthropics/skills)

skill-creator helps you write a new skill, run test prompts against it, compare results and tune its description so it triggers at the right time. mcp-builder is a guide for building MCP servers in Python (FastMCP) or TypeScript, starting from research and tool naming. Both come with example-skills.

Use skill-creator once you notice yourself pasting the same procedure into Claude Code every week. Start with one task, define what a correct result looks like, then test the skill on a few representative prompts. Our guide to writing a Claude Code skill covers the file format, installation paths and testing.

How to judge a skill before installing it

Treat a skill as software you are adding to your development environment. It can include scripts, tool instructions and external dependencies. A marketplace listing or a familiar publisher name does not replace reading what you are about to install.

A short check before you install:

1. Read SKILL.md end to end. The description should match what the body tells Claude to do.

2. Open every script in scripts/ and any other bundled file. Look for network calls, file access outside the project, or commands that don't fit the purpose.

3. Check the allowed-tools frontmatter. The Claude Code docs warn that a skill can grant itself broad tool access, and that repository skills apply it even in folders you haven't trusted.

4. Be wary of skills that fetch content from URLs at run time. Anthropic notes that fetched content can carry instructions, and that a trusted skill can be compromised if its external dependencies change.

5. Install plugins from a session. In Claude Code, /plugin install opens the plugin's details pane first, and the plugin docs say to read it because a plugin can run hooks and MCP servers.

Install one skill for a task you repeat, run it on a small change, and inspect the result before adding a larger pack. In supported Claude Code versions, /skill-doctor reports context cost and unused skills. For a review workflow you can inspect line by line, use our copyable Claude Code review skill. For marketplaces, installation scopes and team-wide distribution, see our Claude Code plugins guide.

Awesome lists for finding more

Curated lists are useful for discovering more skills, but a listing does not establish that every linked repository has been audited. Check the individual skill’s publisher, license and bundled files.

• travisvn/awesome-claude-skills: a curated list of Claude skills, resources and tools.

• ComposioHQ/awesome-claude-skills: a large list of skills grouped by category.

• hesreallyhim/awesome-claude-code: skills, agents, hooks, status lines and plugins for Claude Code.

• VoltAgent/awesome-agent-skills: official skills from development teams plus community skills, in the cross-agent format.

• alirezarezvani/claude-skills: one large library of skills and plugins covering engineering, DevOps and non-coding work.

FAQ

What are the best Claude skills for beginners?

Start with one built-in skill for a task you already do: /verify to exercise a change or /code-review to inspect a diff. Add frontend-design or webapp-testing when you need that specific workflow. Try a planning pack such as Superpowers after you decide you want its broader process.

Do Claude Code skills work in Codex, Cursor or other agents?

Many use the shared Agent Skills format, but scripts, tool access and installation paths can depend on the agent. Read the publisher’s instructions for your target and try a small task after installing. Claude Code’s bundled skills, such as /verify, are specific to Claude Code. Our Codex skills guide and Cursor skills guide cover those agents’ locations and invocation rules.

Where does Claude Code store installed skills?

Personal skills live in ~/.claude/skills/<skill-name>/SKILL.md and load in every project. Project skills live in .claude/skills/ in the repository; commit them and everyone on the team gets them. Plugin skills load wherever the plugin is enabled, and you call them as /plugin-name:skill-name.

Are Claude skills safe to install?

A skill can include executable scripts and instructions to use tools, so review the full package and install only what you intend to use. Claude Code’s disableSkillShellExecution setting blocks inline shell expansion from user, project and plugin skills. It does not sandbox a skill or block every command the agent might run through its tools.