$ man help
Frequently asked questions
Quick answers to the questions we hear most often. Still stuck? Drop us a line — humans reply within a working day.
// Basics
$What is a Claude skill?
A Claude skill is a packaged set of instructions, context, and — optionally — tools that turn a general-purpose model into a specialist. Skills are defined by a human-readable SKILL.md file and distributed through ClaudeSkill so anyone running Claude can install one with a single command.
$Is ClaudeSkill affiliated with Anthropic?
No. ClaudeSkill is an independent, community-run marketplace. Claude and Anthropic are trademarks of Anthropic, PBC; we use them strictly to describe compatibility. Read the full disclaimer on our /disclaimer page.
$Does ClaudeSkill cost anything?
Browsing, installing, and publishing open-source skills is free. We may introduce optional paid tiers for private skills and team workspaces in the future, but the public marketplace will remain free.
// Installing
$How do I install a skill?
Open the skill page and copy the install command — for example, `npx @claudeskill/install owner/skill`. Paste it into any shell where Claude Code is configured. The command fetches the SKILL.md, validates its signature, and registers the skill with your local Claude CLI.
$Do I need Claude Code to use a skill?
Claude Code is the fastest path, but skills are plain markdown — you can also paste a SKILL.md into any Claude-compatible agent (Cursor, Raycast, Warp, the Anthropic API directly) and it will read as instructions.
$Can I install a skill in a specific project only?
Yes. The install command supports a `--scope=project` flag that writes the skill into `.claude/skills/` in the current repo. Skills installed this way travel with the repository and are shared with teammates via version control.
// Publishing
$How do I publish my own skill?
Create a SKILL.md file with a short description, triggers, and instructions. Push it to a public GitHub repo, then click 'Submit a skill' and paste the repo URL. Our scraper reads the manifest, validates it, and a human reviewer publishes it within 24 hours.
$What does the review process check for?
We check that the skill (a) does what it claims, (b) does not exfiltrate data or run destructive shell commands by default, (c) declares a real license, and (d) does not impersonate another creator. We reject skills that are spam, paywalled behind opaque services, or that violate our content policy.
$What license should I use?
We recommend MIT or Apache-2.0 for skills that wrap publicly-available techniques. If your skill bundles original copy, images, or datasets, CC-BY-4.0 may be more appropriate. Every skill must declare its license in SKILL.md or package.json.
$Can I update a published skill?
Yes — push a new commit to your GitHub repo and our scraper will re-index within an hour. Bump the version in SKILL.md so installers know to refresh. We keep a changelog of the last 10 versions.
// Safety
$How do I report a malicious or broken skill?
Every skill page has a 'Report' link at the bottom of the sidebar. Reports are triaged within 24 hours; clearly-malicious skills (credential exfiltration, destructive shell commands) are unpublished immediately.
$What runs on my machine when I install a skill?
The install CLI downloads one or more markdown files, verifies their checksum against the registry, and writes them into your Claude config. Skills do NOT execute arbitrary code at install time — they are read by the agent when you invoke them.
// Business
$Can I advertise my product on ClaudeSkill?
Yes — we run a small set of ad slots on high-traffic pages. Reach out via /contact to discuss placements. We do not accept ads that impersonate skills or mislead users about what they install.
$Do you share user data with advertisers?
No. We do not sell or share individual user data. Ad slots are served contextually (based on the page category, not user profile). See /privacy for the full data-handling policy.
Still have questions?
We're a small team, and we read every message. The fastest way to get help is /contact.