Best Claude Code Skills in 2026: 20 Slash Commands Worth Installing

Claude Code ships with a handful of built-in skills. The real power comes from custom ones — .md files that give Claude structured workflows for specific tasks. Some are team-shared, others personal productivity hacks.

Here are 20 worth installing, organized by what they do.

How Skills Work (30 seconds)

A skill is a markdown file placed in .claude/commands/ (project-level) or ~/.claude/commands/ (personal). The filename becomes the /slash-command. The file content is the prompt Claude receives when you invoke it.

---
name: review
description: Review code changes for bugs, security issues, and style
---

Review all staged changes. For each file, check for:
1. Logic errors and edge cases
2. Security vulnerabilities (injection, XSS, auth bypass)
3. Performance issues (N+1 queries, unbounded loops)
...

Type /review and Claude follows those instructions with your actual code.

You can also install skills from URLs:

claude skill add --url https://raw.githubusercontent.com/user/repo/main/skill.md

Git & Workflow

1. /commit (built-in)

Generates a commit message from staged changes, follows your repo's commit style. Built into Claude Code — no install needed.

2. /create-pr

Creates a pull request with a structured title, summary, and test plan. Reads the full diff, writes the description, pushes the branch, and opens the PR via gh.

Source: wshobson/commands

3. /commit-fast

Skip the interactive commit flow. Stages all changes, generates a message, and commits in one shot. For when you trust Claude's judgment and want speed.

4. /rewind

Safety net. Reverts both the conversation and file changes to a previous checkpoint. Essential when Claude goes down a wrong path — undo everything in one command.

Code Quality

5. /review (built-in)

Reviews changed code for bugs, security issues, and style violations. Built-in, but you can override it with a custom version that knows your team's conventions.

6. /simplify (built-in)

Scans recently changed code for unnecessary complexity. Finds duplicate logic, premature abstractions, and over-engineering. Then fixes them.

7. /security-audit

Deep security review: injection vectors, auth bypass, secrets in code, dependency vulnerabilities. More thorough than /review — focused exclusively on security.

Source: awesome-claude-code

8. /test-coverage

Identifies untested code paths and generates tests for them. Reads your existing test patterns and matches the style (Jest, Vitest, pytest, etc.).

Architecture & Planning

9. /plan (built-in)

Enters plan mode — Claude explores the codebase, identifies affected files, and proposes an implementation plan before writing code. Use this for any non-trivial change.

10. /architect

Like /plan but focused on system design. Given a feature request, Claude maps the data flow, identifies affected components, considers tradeoffs, and proposes the architecture before any code is written.

11. /refactor

Refactor a specific module with a goal: "extract this into a service," "split this component," "migrate from callbacks to async/await." Claude reads the full context, plans the refactor, then executes.

Documentation

12. /document

Generates documentation for a file, module, or function. Reads the code, infers the intent, and writes docs that match your project's style (JSDoc, docstrings, README sections).

13. /changelog

Reads the git log since the last tag/release and generates a structured changelog. Groups changes by type (features, fixes, breaking changes).

Testing

14. /test

Generates tests for a specific file or function. Reads the implementation, identifies edge cases, and writes tests in your project's testing framework.

15. /e2e

Writes end-to-end tests for a user flow. "Test the signup flow: register, verify email, first login, onboarding." Claude generates Playwright or Cypress tests.

Productivity

16. /compact (built-in)

When your conversation gets long and Claude starts losing context, /compact compresses the history. You can specify what to preserve: "/compact keep the auth refactor decisions."

17. /loop

Runs a command or workflow on a recurring interval. "/loop 5m /review" checks your code every 5 minutes. Useful for long-running tasks where you want periodic Claude input.

18. /batch

Applies a prompt across multiple files. "Add error handling to all API routes" runs on each file matching a pattern. Useful for bulk refactors.

Domain-Specific

19. /recommend-tool

Uses the Toolradar MCP server to find, compare, and recommend software tools. Searches 8,400+ tools, compares the top 3, and gives a structured recommendation with top pick, runner-up, and budget option.

claude skill add --url https://raw.githubusercontent.com/Nadeus/toolradar-mcp/main/skill-recommend-tool.md

Then: /recommend-tool "best free CRM for a 5-person startup"

Requires the Toolradar MCP server to be connected. Setup guide →

20. /deploy

One-command deployment with pre-flight checks. Runs tests, checks for uncommitted changes, builds, deploys, and verifies the deployment succeeded. Customize for your stack (Vercel, AWS, Docker, etc.).

Where to Find More

• awesome-claude-code — curated list of skills, hooks, and plugins
• wshobson/commands — production-ready slash commands
• Claude Code docs — official documentation on creating skills

Build Your Own

The best skills are the ones specific to your project. Common patterns:

• Team conventions: /style-guide that enforces your naming conventions, file structure, and patterns
• Onboarding: /explain-codebase that walks new developers through the architecture
• Ops: /incident-response that pulls logs, checks monitoring, and suggests fixes
• Content: /write-blog-post that follows your editorial checklist and SEO guidelines

A skill is a prompt. If you find yourself giving Claude the same instructions repeatedly, extract them into a skill.

Want your AI to recommend software tools? Install the /recommend-tool skill →

Need the MCP server behind it? Set up Toolradar MCP in 2 minutes →