Recommendations · 4 min read
The Best Claude Code Skills in 2026, by Use Case
The Claude Code skills worth installing first, grouped by job: debugging, code review, frontend, databases, security, DevOps, documents and AI apps.
The best Claude Code skills are the ones that change what Claude does on a task you repeat, not the ones that restate what it already knows. For most developers that means a small core of workflow skills that improve every session, plus a handful matched to your stack. This page lists the picks we would install first, grouped by the kind of work they cover, with a sentence on why each one earns its place.
Everything here is free, MIT licensed, and installs with one command:
curl -fsSL https://claudeskills.page/install.sh | sh -s -- systematic-debugging code-review
The core five: install these on every machine
These skills change Claude’s behavior on almost any codebase, regardless of language or framework.
| Skill | What changes |
|---|---|
| Systematic Debugging | Claude reproduces the bug and explains the mechanism before editing anything, and finishes with a regression test |
| Task Decomposition Plans | Large requests become a written plan with checkpoints before code is touched |
| Code Review | Findings ranked by severity with concrete fixes, minus the nitpicks a linter should catch |
| Commit Craft | Atomic commits whose messages explain why, so history stays bisectable |
| CLAUDE.md Architect | A short memory file Claude actually follows, with task-specific knowledge moved into skills |
If you only install one collection, make it the Starter Pack, which bundles these with TDD, pull request writing and context hygiene.
Best skills for getting more out of Claude Code itself
Power users get the biggest return from skills that teach Claude to configure and extend itself:
- Skill Authoring turns a prompt you keep retyping into a reusable skill with the right trigger description.
- Hooks Automation converts “always do X” rules into hooks that run every time, such as formatting after each edit or blocking dangerous commands.
- Subagent Design creates scoped subagents for research, review or tests, with limited tools and their own context.
- MCP Server Builder builds and tests a Model Context Protocol server so Claude can reach your internal systems.
- Headless CI Agent runs Claude Code non-interactively in CI for review and triage.
The Claude Code Power User collection adds permission policies, worktree parallelism and the Agent SDK.
Best skills for frontend development
- React Patterns and Next.js App Router keep server and client boundaries, caching and effects correct for current versions.
- Tailwind CSS covers the v4 CSS-first configuration that trips up older habits.
- Accessibility Audit runs automated checks first, then the manual WCAG 2.2 passes tools cannot judge, and writes findings engineers can act on.
- Core Web Vitals diagnoses LCP, INP and CLS from field data instead of guessing from a lab score.
See the Modern Frontend Stack and Web Performance Kit collections for the full set.
Best skills for backends and databases
- REST API Design produces consistent resources, status codes, pagination and RFC 9457 errors.
- Postgres Schema Design and Postgres Indexing work from real query plans rather than intuition.
- Zero-Downtime Migrations uses the expand-and-contract pattern so deploys never lock a busy table.
- Background Jobs and Webhook Handling build in idempotency, retries and signature checks from the start.
- Pagination and Filtering and API Versioning keep list endpoints fast and let public APIs change without breaking clients.
The Production Backend collection puts these in adoption order.
Best skills for security reviews
Security skills on this site cover defensive work and explicitly authorized testing only.
- OWASP Top 10 Review checks code category by category and reports findings with severity and fixes.
- Threat Modeling produces a STRIDE model with trust boundaries before you write the risky code.
- Secrets Management finds leaked credentials and sets up rotation and vaults.
- Supply Chain Audit triages dependency vulnerabilities and lockfile hygiene.
- MFA Implementation and SAML SSO cover the sign-in features enterprise customers ask for first.
Start with the Security Review Kit.
Best skills for DevOps and shipping
- Dockerfile Optimization builds small, cache-friendly, non-root images.
- GitHub Actions speeds up workflows and replaces stored cloud keys with OIDC.
- Terraform IaC reviews plans for destructive changes before apply.
- Incident Response keeps a live incident organized: mitigation first, a clear timeline and regular updates.
Bundle them with the Ship to Production and Cloud Native Ops collections.
Best skills for documents: PDF, Excel, Word and PowerPoint
Office files are where general-purpose agents most often produce something that looks right and is broken underneath.
- PDF Toolkit extracts text and tables, fills forms, and merges or splits files.
- Excel Spreadsheets writes workbooks whose formulas still recalculate in Excel.
- Word Documents edits DOCX files without destroying their styles.
- PowerPoint Decks builds decks from notes and checks for overflow.
The Office Documents collection adds OCR, true redaction and pandoc publishing.
Best skills for building AI features
- LLM App Builder integrates a model API with streaming, retries and cost control.
- LLM Evals builds an evaluation set before you tune prompts, so changes are measured.
- RAG Pipelines covers chunking, hybrid retrieval, reranking and citation checks.
- Prompt Injection Defense designs guardrails for apps that read untrusted content.
- LLM Gateway Routing adds fallbacks, model tiering and per-route cost tracking once you call more than one model.
See the AI Engineering collection.
Beyond code: product, research and automation
Claude Code is useful well beyond writing software. The Product Manager Toolkit covers discovery, user stories and launch readiness. The Research Assistant runs literature reviews and fact-checking with sources that exist. Automate Everything turns repetitive chores into scripts, scheduled jobs and n8n workflows.
How we chose these
Every skill on claudeskills.page is written to the same editorial standard: a selective trigger description, a real workflow, decision tables with numbers, known pitfalls and a verification step. The picks above are the ones that changed Claude’s output most visibly in everyday use. Browse the full directory to find skills for your exact stack, or read how to install skills if you are new to them.
Frequently asked questions
How many Claude Code skills should I install?
As many as match your work. Claude Code keeps only each skill's short description in context, roughly 100 tokens per skill, and loads the full instructions on demand. Twenty well-chosen skills cost about 2,000 tokens of standing context, far less than a long CLAUDE.md.
Should I install skills globally or per project?
Install general workflow skills such as systematic-debugging, code-review and commit-craft globally in ~/.claude/skills/. Install stack-specific skills, like a framework or database skill, in the repository's .claude/skills/ folder and commit them so every teammate gets the same behavior.
Do skills conflict with each other?
Rarely, because each skill describes a different trigger. If two skills both claim the same request, Claude picks the closer description. If one keeps firing when you meant the other, tighten the descriptions; the skill-description-tuning skill walks through it.
Are these skills safe to install?
Every skill on this site is plain Markdown plus, for some, small scripts you can read on the skill page before downloading. Read any script before running it, as you would with any code from the internet.