There are a lot of Claude Code Skills floating around — bundled ones, official marketplace plugins, community catalogs, and repos full of SKILL.md files someone exported from a side project. Most of them are not worth the slot in your head, let alone your repo. A skill earns its place when it encodes a process you keep re-explaining. Everything else is noise.
So this is not a foggy “top 50” list. It starts with the actual bundled skills you already have, then the installable skill plugins worth checking, with links and install commands. New to the format? Start with Claude Code Skills explained.
The test to apply throughout: if you find yourself pasting the same instructions into Claude more than twice, that’s a skill. If you don’t, it isn’t.
The actual shortlist
Start here before browsing random repos.
2026 shortlist
Skills and skill plugins worth using
12 skills
/code-review
Use after edits when you want a real review pass instead of a vague check.
Install
Bundled, no install Docs / source
Why it is worth using: It gives review a defined prompt and makes Claude look for defects, regressions, and missing safeguards with a clearer job.
/debug
Use for failed tests, stack traces, and reproducible bugs.
Install
Bundled, no install Docs / source
Why it is worth using: It turns debugging into a procedure: reproduce, inspect evidence, form a hypothesis, and verify the fix.
/run and /verify
Use when Claude needs to launch the app and prove the change works.
Install
Bundled, no install Docs / source
Why it is worth using: It forces verification against a running app instead of ending with 'the code looks right.'
/run-skill-generator
Use once when your app has a non-obvious launch or verification recipe.
Install
Bundled, no install Docs / source
Why it is worth using: It records your project's real run process so future /run and /verify calls stop rediscovering it from scratch.
/loop and /batch
Use for iterative fixes, repetitive edits, and repeated checks.
Install
Bundled, no install Docs / source
Why it is worth using: They help Claude keep a tight work loop instead of doing one pass and stopping too early.
/claude-api
Use when building against Claude APIs.
Install
Bundled, no install Docs / source
Why it is worth using: It keeps API work closer to current Claude conventions instead of relying on generic SDK guesses.
skill-creator
Use when you want Claude to help design, improve, evaluate, and benchmark your own skills.
Install
/plugin install skill-creator@claude-plugins-official Docs / source
Why it is worth using: It is the best starting point once your useful workflows are team-specific and worth packaging properly.
commit-commands
Use when diff summaries, commit messages, pushes, and PRs are repeated work.
Install
/plugin install commit-commands@claude-plugins-official Docs / source
Why it is worth using: It turns the end-of-task git routine into repeatable commands, which keeps commits more consistent.
frontend-design
Use when frontend output keeps drifting into generic AI UI.
Install
/plugin install frontend-design@claude-plugins-official Docs / source
Why it is worth using: It packages practical frontend design guidance around layout, hierarchy, states, and polish.
feature-dev
Use when you want a repeatable explore-design-implement-review flow.
Install
/plugin install feature-dev@claude-plugins-official Docs / source
Why it is worth using: It helps feature work follow a real sequence instead of jumping straight from idea to edits.
superpowers
Use for brainstorming, debugging, TDD, review, and skill authoring workflows.
Install
/plugin install superpowers@claude-plugins-official Docs / source
Why it is worth using: It is broad and process-heavy, so read it first, but it is one of the few broad workflow packs worth trying on a small task.
Platform skill packs
Use only when the platform matches your actual stack.
Install
Install only the pack you use: terraform, aws-core, azure, cloudflare, mongodb, prisma, or neon Docs / source
Why it is worth using: They beat generic infra advice because they bundle platform-specific guidance and tools, but they are dead weight when they do not match your work.
To install any marketplace skill plugin, use the same pattern:
/plugin install skill-creator@claude-plugins-official
/reload-pluginsBundled skills worth using first
The bundled skills are the easiest yes because they are already available.
/code-review and /debug are the two daily drivers. Review gives Claude a sharper job after it changes code; debug gives it a procedure when something fails. /run and /verify are the quality gate pair: they ask Claude to launch the app and confirm behavior. If your project has a non-obvious launch process, run /run-skill-generator once so Claude records the real recipe in .claude/skills/.
Code-review and pull-request skills
This is the category most teams should build first. A review skill encodes the checklist your senior engineers apply by reflex: security checks, error handling, naming, test coverage, the specific traps in your codebase.
What it standardizes: the difference between “looks fine” and a real review. Instead of a generic pass, Claude applies your rules — “never log secrets,” “every new endpoint needs an auth check,” “no raw SQL string interpolation.” Use it on every PR. The payoff scales with team size, because it makes junior and senior reviews land in the same place.
Frontend and design-guideline skills
If you have a design system, a frontend skill keeps Claude inside it. Without one, an agent will happily invent spacing, reach for the wrong component, or hardcode a color that should be a token.
What it standardizes: component choices, spacing scale, accessibility requirements, when to use which layout primitive. Point the skill at your design tokens and component library and it stops freelancing. Use it on any project with an established UI kit. Skip it for throwaway prototypes where consistency doesn’t matter yet.
Testing skills
A testing skill encodes how your team writes tests — the framework, the file layout, what a good test name looks like, what to mock and what to leave real.
What it standardizes: structure and coverage expectations. Claude can write tests without a skill, but it’ll guess your conventions. A skill that says “use the table-driven style, name tests Test_<unit>_<condition>, mock the network layer but not the database in integration tests” produces tests that match the rest of the suite. Use it whenever you’re adding features that need tests, which is most of the time.
DevOps and infrastructure skills
For infra work, a skill keeps Claude from generating plausible-looking config that doesn’t match your environment. Docker, Kubernetes, Terraform, CI pipelines — each has conventions that a skill can capture.
What it standardizes: your base images, your naming, your tagging strategy, the guardrails before anything touches a cluster or runs an apply. This category is broad enough to deserve its own treatment — see best Claude Code Skills for DevOps for specific examples and the description lines that make them trigger. Use these when you manage real infrastructure and want the agent to respect it.
Memory and meta skills
A different kind of skill: ones that shape how Claude works rather than what it produces. A “planning” skill that forces a task breakdown before coding, a “summarize-session” skill that writes a clean handoff note, a skill that enforces your documentation format.
What they standardize: process. These are easy to overdo — a meta skill that fires on everything just adds friction. Keep them narrow and tied to a clear trigger. Use them when you have a workflow step you keep forgetting to do by hand.
How to judge any skill before you keep it
The community has produced large catalogs of skills across public repos. Volume is not quality. Most of them solve a problem you don’t have, or solve it in a way that doesn’t match how you work. Before you install or keep one, run it through a quick filter.
Keep it or drop it
| Repeatable | Does it encode a process you do more than twice? If not, skip. |
|---|---|
| Distinct trigger | Is the description specific enough to fire on the right task — and not overlap with another skill? |
| Readable | Have you read the SKILL.md? You're trusting it with instructions Claude follows. |
| Yours where it matters | Team conventions belong in skills you write, not borrowed ones. |
The short version
Use the bundled skills first: /code-review, /debug, /run, and /verify cover a surprising amount of daily work. Install skill-creator when you want to build your own skills, commit-commands for git workflow, and stack-specific packs only when they match your tools. Build your own for code review, testing, and anything tied to your team’s conventions — only you can write those accurately. Borrow community skills sparingly, and only after reading them.
The honest takeaway: a small set of sharp, well-described skills beats a folder full of borrowed ones you never see fire. When you’re ready to build, follow how to write a Claude Code Skill, and check the DevOps-focused roundup if you live in infra.
