The Best Claude Code Skills to Install in 2026
The best Claude Code skills, ranked: Superpowers, GSD, Claude-Mem and more. How they work, how to install them, and which ones to skip.
What are the best Claude Code skills to install right now, and which ones should you skip?
The best Claude Code skills extend Claude with reusable expertise you can install in minutes: community favorites like Superpowers, GSD, and Claude-Mem, plus Anthropic’s official repository. They rank high because they are tested, maintained, and solve real recurring work. Install the ones that match your stack, keep the shortlist small, and skip everything you cannot audit.
Most skill lists are favorites from developers who never audit what they recommend. This is the plain version. I run a team of AI agents where every agent carries its own skill file, and this post is that pipeline’s output.
TL;DR
- A Claude Code skill is a folder with a SKILL.md file that teaches Claude one repeatable job. You install it by copying the folder into ~/.claude/skills/ or .claude/skills/.
- Start with Superpowers, GSD, Claude-Mem, Ralph Loop, and the claude-skills collection, plus Anthropic’s official repository for document work.
- Keep the shortlist at five to seven. Match each skill to a job you do weekly. Skip anything you cannot read and audit in five minutes.
What are the best Claude Code skills?
The best Claude Code skills right now are Superpowers, GSD, Claude-Mem, Ralph Loop, and the claude-skills collection, plus Anthropic’s official repository for document work. Developer forum threads name these five again and again. They win because they are maintained, auditable, and built for the jobs you repeat weekly.
A skill is a directory with a SKILL.md file at its root. Claude preloads only the name and the description at startup, roughly 100 tokens per skill, and reads the full file when a request matches, per the Claude Code skills documentation.
| Skill | Category | Source |
|---|---|---|
| Superpowers | Planning and execution sequence | obra / superpowers, community |
| GSD | Task management | TÂCHES / gsd-build, community |
| Claude-Mem | Session memory | thedotmack, community |
| Ralph Loop | Loop iteration | community |
| claude-skills | Community index of skills | topic collection |
| Official document skills | PDF, Word, Excel, PowerPoint work | official skills repository |
The five community picks come up consistently in developer forum threads.
Which skill should I install first?
Install the skill that matches the biggest job you repeat every week. Code daily? Start with Superpowers and its clarify, spec, plan, execute, review sequence. Documents? Start with Anthropic’s official document skills. Losing context between sessions? Start with Claude-Mem.
That sequence is what makes Superpowers the usual first pick, per this top-ten list.
What about the other skills on the lists?
Most are real. frontend-design, vercel-react-best-practices, spartan-ai-toolkit, mattpocock/skills, planetscale, web-design-guidelines, doc-coauthoring, code-simplifier, TerraShark, Grill Me, and Context Mode all do honest work for the right person. None beats a skill matched to your actual stack, which is why the shortlist method matters more than the names.
Those names come from the two pages that rank for this query: a personal top ten and a tested shortlist that tried 33 skills and kept six (welcome developer, artificial corner).
The skill that wins is the one matched to the work you actually do, not the most-shared list.
How do I install a Claude Code skill?
To install a skill, copy its folder into ~/.claude/skills/ for personal use, or into .claude/skills/ for a single project. Skills are filesystem based, so copying a folder is the whole install, and custom commands have merged into skills so both paths create the same command.
Where do personal and project skills go?
Personal skills go in ~/.claude/skills/ and apply to every project. Project skills go in .claude/skills/ at the project root and apply only there. The folder name becomes the command, so summarize-changes runs as /summarize-changes. Copying a folder is the entire install.
The paths above come from the custom skills section of the Claude Code docs.
How do slash commands fit in?
Custom commands have merged into skills. A file at .claude/commands/deploy.md and a skill folder at .claude/skills/deploy/SKILL.md both create /deploy and behave the same way. Existing command files keep working, but skills add supporting files, which is why they are recommended now.
That merge is confirmed in the slash commands docs.
How do I know a skill is safe to install?
You do not trust a skill, you audit it. Read every bundled file, check code dependencies, and look for instructions that point Claude at external network sources. Malicious skills can introduce vulnerabilities in the environment where they run, or direct Claude to exfiltrate data, so install only from trusted sources.
Anthropic spells out the risk and the audit steps in its Agent Skills engineering post. Most tutorials for installing claude code skills leave this step out, and it is the step that matters most.
I install skills the way I check a tool before paying for it. Read the file, run it once on real work, delete it if it does not earn its place. The automation that changed my week was putting follow-up on autopilot, and a skill is that same instinct in file form.
Copy the folder, read the SKILL.md, test it once. That is the entire install process.
How do I pick a shortlist that fits my stack?
Pick skills by workload, not by popularity. List the jobs you repeat weekly, match one skill to each job, and keep the shortlist under seven. Delete anything that does not fire on real work, and skip any skill whose file you cannot read in five minutes.
What should I skip?
Skip anything you cannot audit in five minutes, anything from an unknown source, and anything whose trigger fires on work you do not do. Apply the trust rules from the install section first, then the stack rules. A skill that fails either group does not earn a slot.
What is the three-check audit?
Three checks decide every install. One, does it match a job you did by hand last week? Two, can you read the whole SKILL.md in five minutes? Three, does the trigger description fire only when you want it? Fail one check, and the skill goes back to the list.
| Check | Pass | Fail |
|---|---|---|
| Matches a weekly job | You did this job by hand last week | You installed it because a post ranked it |
| Readable in five minutes | You can explain the SKILL.md in one sentence | You skipped the file and trusted the stars |
| Trigger stays quiet | It fires only when you ask | It fires on half your sessions |
I run my own outreach system, and it lands about 200 form submissions a day. At that volume, one rule decides every tool I keep: I have to be able to inspect it. The same rule picks my skills.
It works for claude code skill lists the way it works for everything else. The only skill list that matters is yours. The same thinking applies to AI agents for small business: install the workflow, not the hype.
Seven skills or fewer, each matched to a job you do weekly, each readable in five minutes.
How does a SKILL.md file actually work?
A skill is a folder holding a SKILL.md file. Claude preloads only the name and description, about 100 tokens, and reads the rest when matched. The name must be lowercase, at most 64 characters, with hyphens. The description states what the skill does and when to use it.
What goes in the frontmatter?
The YAML frontmatter is the skill’s identity. name must be lowercase, at most 64 characters, with hyphens, and no reserved words like anthropic or claude. description states what the skill does and when to use it. allowed-tools pre-approves the tools a skill may use.
The field rules come from SKILL.md frontmatter in the official docs.
What is progressive disclosure?
Claude loads the skill in stages. The name and description load at startup, about 100 tokens per skill. The body loads only when the skill triggers. Reference files and bundled scripts load only when needed, and scripts run through bash so only their output enters context.
That staged loading is the Claude Code skills documentation’s “progressive disclosure” design. Skills sit alongside hooks and subagents in the same configuration system.
How do skills relate to MCP?
MCP connects Claude to external tools and services. Skills teach the workflows around those tools: when to call them, in what order, and what to check. A skill and an MCP server do different jobs, and they work better together.
Anthropic positions skills and MCP as complements in its Agent Skills engineering post.
Do skills ship real work?
Yes. I built the AI SEO team that wrote this page, and every agent in it carries a SKILL.md: researcher, planner, copywriter, reviewer, and publisher. The 19-skill bundle I sell ships Skill Creator and Anti-Slop Humanizer in the same format. Both are plain folders with a SKILL.md at the root.
The skills bundle is the same skill format, packaged for a buyer instead of a codebase, and it is the proof that the format carries production work.
AutomateReal skills
The description is the trigger. Write it to say what the skill does and when to use it, because that is how Claude decides to load the file.
How do I write my own Claude Code skill?
Create a folder, add a SKILL.md, and write a name under 64 characters plus a description that states what the skill does and when to use it. Then write short instructions in the body. Start with one job and one trigger. Test it on real work before you trust it.
Where do I even start?
Start with the frontmatter, then keep the body tight. If a skill needs reference material, split it into bundled files, because Claude reads those only when needed. If it needs tools, declare them in allowed-tools so it requests only what it uses.
What makes a skill fail?
The failures are never dramatic. Bloated prompt walls, trigger descriptions so broad they fire on half your sessions, scripts nobody can read, or instructions that quietly call out to the network. A skill that grabs context it does not need slows every task it touches.
How do I test a skill before I trust it?
Run it on last week’s job, not a toy example. My test is one sentence: it has to do the job better than I did by hand, and its trigger has to stay quiet when it should. Skills that pass stay. Failures get rewritten or deleted within a week.
A skill is also the closest thing to a written job description for your first AI employee. Write it once, and the same standard applies to every session.
Write the description for a machine, then prove the body on real work.
Related: what are claude code skills
Related: sell claude skills (make money with claude skills)
Related: claude skills vs mcp vs agents
Related: claude code skills for marketing
Related: AI agents for dental practices
Related: AI agents for insurance agencies
FAQ
What is the difference between a skill, a prompt, and a slash command?
A prompt is a one-off instruction you type. A skill is a file that teaches a complete, repeatable workflow. A slash command is the same thing as a skill now: custom commands have merged into skills, so both create the same command and behave the same way.
Where are Claude Code skills stored on disk?
Personal skills live in ~/.claude/skills/ and apply to all your projects. Project skills live in .claude/skills/ at the project root and apply to one project. Both are plain folders, so you can back them up, version them with git, and share one by copying the folder.
Do skills work with Claude Code subagents?
Yes. Skills run in subagents, and Claude Code extends the Agent Skills standard with subagent execution. A skill can hand part of a job to a subagent, which keeps the main thread short and the context window clean.
How do I write a skill that calls a tool or a script?
Write the call into the body of the SKILL.md and list the tools in the allowed-tools frontmatter field, so the skill requests only what it needs. Skills can bundle scripts that run through bash, and only their output enters context. Test the skill against the real tool, not a mock.
Should I install skills from GitHub or build my own?
Start with trusted sources such as Anthropic’s official repository, then build your own for the work that repeats in your business. A skill you write for your own workflow beats a famous one that does not fit, and you can audit every line of it.
What should you install today?
Install one skill this week for one job you did by hand last week. Read it, run it once, keep it only if it earns the slot. Skills are the cheapest automation you will install this year.
The format is young, and anyone selling a decade of skill expertise is selling you something. If you want help finding that first workflow, a discovery call maps it in about thirty minutes.
AutomateReal services