Stacktree
By · Founder, Stacktree · Last updated
npx skills

One command puts a skill in every agent you use.

npx skills is the CLI for the open agent skills ecosystem. A skill is a folder with a SKILL.md instruction file. The CLI fetches one from a repo and wires it into Claude Code, Codex, Cursor, OpenCode and 75 more agents at once. Nothing to install first: npx fetches the CLI too.

Get started free

No card · 3 pages free · about a minute

What is npx skills?

npx skills runs the skills CLI, an open-source tool from Vercel Labs for installing agent skills — folders containing a SKILL.md instruction file that an agent loads when a task matches. One command fetches a skill from a Git repo and wires it into every coding agent you have, so the same instructions work in Claude Code, Codex, Cursor and OpenCode without being rewritten for each. It is MIT-licensed, and because npx fetches it on demand there is nothing to install beforehand.

The commands

Seven, and you will mostly use the first two.

  • npx skills add <source> — install a skill and wire it into your agents.
  • npx skills use <source> — run one skill without installing it. It writes the files to a temp directory and prints the generated prompt, so npx skills use owner/repo@skill-name | claude works as a one-off. Add --agent claude-code to launch an agent interactively instead.
  • npx skills list (alias ls) — what is installed, project and global.
  • npx skills find [query] — search for skills, interactively or by keyword.
  • npx skills update [skills] — pull the latest version of what you installed.
  • npx skills remove [skills] — take them back out of your agents.
  • npx skills init [name] — scaffold a new SKILL.md to write your own.

Where a skill can come from

The source argument is deliberately loose. All of these resolve:

  • GitHub shorthand — owner/repo, the common case.
  • A full URL, or a path straight to one skill inside a repo.
  • GitLab, or any git URL, including SSH.
  • A local path such as ./my-local-skills, which is how you test one you are writing.
  • Private repositories, using the git credentials you already have. The CLI tries normal git auth first, then GitHub CLI, then SSH.
  • A direct download URL pointing at a SKILL.md or a zip/tar archive.

Where the files land

Two scopes, and the default is the one you usually want:

  • Project (default) — ./<agent>/skills/, committed with the repo so your team gets the same skills.
  • Global (-g) — ~/<agent>/skills/, available in every project you open.

When several agents are installed it symlinks each of them to one canonical copy, so updating a skill updates it everywhere. Pass --copy if your environment cannot follow symlinks.

Useful flags: -a to target specific agents, -s to pick specific skills ('*' for all), -l to list what is in a repo without installing, and -y to skip prompts in CI.

A real one to try it on

Reading about a CLI is less useful than running it. Ours publishes whatever your agent just built to a private URL a person can open, which is a task most agents cannot do at all: 

npx skills add stevysmith/stacktree-skill

Then tell your agent “publish this and give me the link”. It returns an unguessable URL that opens in any browser with no account for the reader. Full disclosure: Stacktree is our product, and the rest of this page is about our own skills. The CLI above is not ours — it works exactly the same with any repo.

publish

Pipe any HTML artifact to a private URL the moment it exists.

Install

Then

"publish this dashboard and give me the link"

The agent pipes the file to stacktr.ee and returns a private, unguessable URL into the conversation. No project, no build, no DNS.

  • ◇ Anthropic Skills format: works in Claude Code, Codex, Cursor, OpenCode, and any agent that reads SKILL.md.
  • ◇ Private by default; the URL is the credential, viewers need no account.
  • ◇ Replace-in-place: revising the artifact keeps the same URL.

connect · mcp

Give your agent the publish_html tool over MCP, with gates and expiry.

Install

Interactive: signs you in and writes the MCP config for Claude Code, Cursor, Codex, or a plain JSON snippet.

Then

› publish_html(dashboard.html)

The MCP server is the richer surface: update_site, set_password, set_email_gate, set_expiry, list_sites, delete_site, and reading viewer feedback with list_feedback, all as tool calls.

  • ◇ One MCP server, every client: Claude Code, Claude.ai, Cursor, Codex, raw HTTP.
  • ◇ Gating as tool calls: passwords, email-domain gates, expiry, burn-after-read.
  • ◇ Remote server at api.stacktr.ee/mcp if you would rather not run anything locally.

migrate

Move a site off Webflow, Framer, or Squarespace, and verify it in private.

Install

Then

/website-builder-migrate https://yourdomain.com/

Seven phases: mirror, consolidate, rewrite URLs, fix the five known breakages (SRI blocks, lazy chunks, double-encoded filenames), verify in-browser, deploy.

  • ◇ Platform-agnostic, with per-builder notes; Wix honestly flagged as static-template-only.
  • ◇ Captures rendered CMS pages that Webflow's paid export omits.
  • ◇ Deploys anywhere; the no-account path publishes a 24-hour private preview for checking against the original before DNS moves.

pay · x402

Let the agent buy its own key and unlocks. No card, no account, no human.

Install

Third-party x402 client skill (any works, e.g. @x402/fetch); point it at our endpoints with a funded Base wallet.

Then

POST https://api.stacktr.ee/provision

Answers 402 with terms; the agent signs a gasless USDC authorization and gets a persistent stk_live_ key for $1. Unlocks (custom domain, higher limits) work the same way. No wallet? POST /pay/sessions prints a QR a human pays by card, and surplus becomes a balance future actions draw silently.

  • ◇ Live on Base mainnet, x402 v1 and v2; the catalog is public at GET /unlock.
  • ◇ Scan-to-pay fallback: one QR scan funds a prepaid balance ($5 to $20).
  • ◇ The whole loop is real: an agent paid us $1 and published a page that is still up.

Set up your agent in one line

Paste this into your agent. It fetches the setup instructions, installs Stacktree, checks the connection, and learns the parts agents usually get wrong: replacing a page in place so a link you already sent keeps working, reading the real expiry instead of promising permanence, and filing work under the client it belongs to. 

Fetch and follow the setup instructions at https://stacktr.ee/prompt.md

Works in any agent that can fetch a URL. Nothing to install first, and the instructions are plain Markdown you can read before you run them.

FAQ

Frequent questions

What is npx skills? +
It is the CLI for the open agent skills ecosystem, maintained by Vercel Labs and MIT-licensed. It installs agent skills — folders containing a SKILL.md instruction file — into the coding agents you already use, so one skill works across Claude Code, Codex, Cursor, OpenCode and around 75 others rather than being rewritten for each. Because it runs through npx there is nothing to install first.
Do I need to install anything to use npx skills? +
No. npx fetches the CLI on demand, so npx skills add owner/repo works on a machine that has Node and nothing else. You can install it globally if you use it constantly, but for occasional use npx is the point.
Where does npx skills install files? +
Project scope by default, at .//skills/, so the skills are committed with your repo and your team gets them. Pass -g for global scope at ~//skills/, available across every project. With several agents installed it symlinks each one to a single canonical copy so an update propagates everywhere; --copy makes independent copies instead.
Which agents does npx skills support? +
Claude Code, Codex, Cursor and OpenCode are the headline four, with around 75 more including Cline, Continue, Gemini CLI, GitHub Copilot, Goose, Kilo, Roo, Warp, Windsurf and Zed. The -a flag targets specific ones; --agent '*' installs to all of them.
Can I run a skill without installing it? +
Yes, that is what npx skills use does. It resolves the source the same way, writes the files to a temporary directory, and prints the generated prompt to stdout, so npx skills use owner/repo@skill-name | claude pipes it straight into an agent. Add --agent to launch a supported agent interactively with that prompt instead.
What is an agent skill? +
A folder with a SKILL.md instruction file (plus optional scripts) that an agent loads when the task matches. The format originated with Anthropic Skills and is now read by Claude Code, Codex, Cursor, OpenCode, and others, so one skill works across agents. Install is one command: npx skills add owner/repo.
Skill or MCP server: which should I install for publishing? +
Either works; they differ in surface. The skill is the lightest path: instructions plus a publish script using your API key. The MCP server exposes the full toolset (update_site, passwords, email gates, expiry, listing, deletion, reading and resolving viewer feedback) as native tool calls. If your agent speaks MCP, the server is the better default; the skill suits restricted environments.
Can an agent set this up with no human involved? +
Yes, end to end. Skills install from public repos without auth. The API key normally requires a human dashboard login, but an agent with a funded Base wallet can buy its own persistent key for $1 over x402, and an agent without a wallet can print a QR code its human pays by card in two taps. Both paths are live in production.
Are the skills free? +
Yes. Both Stacktree-maintained skills are open source on GitHub, and the third-party x402 client skill is too. Hosting has a free plan (3 pages in total at 25 MB each, live for 7 days); paid actions are a-la-carte ($1 to provision a key, $5 to make one page permanent, $5 / 30 days for a custom domain, $25 / 30 days for higher limits).
Keep reading

Related guides

References

Sources and further reading

Three abilities, one command each.

Install the publish skill, point your agent at an artifact, get a private link back. Free tier, no card; the agent can even pay for itself.

Sign up free →