Home / Agent skills / Claude Code skills
Claude Code skills

Claude Code skills, plus a hosted execution layer

Claude Code skills are SKILL.md folders on your filesystem that teach Claude Code a task. Connect UniversalBench and a skill can also run code in a hosted sandbox, reach your database and the live web, and make validated commits, all through one MCP URL. Your skills folder stays exactly where it is.

Skills are the recipe. The work still has to run somewhere.

A skill is a folder with a SKILL.md file: instructions, and often a few small scripts, that teach your AI how to do a job the same way every time. That is the recipe, and a recipe is not a kitchen. The moment a skill needs to run code, read a database, call the live web, or commit to a repo, that work has to execute somewhere, and you want it to run safely. MCP is how your AI reaches those tools. UniversalBench is the hosted place the work actually runs. The skill is the recipe, MCP is the wire, and UniversalBench is the kitchen.

See it in action

# A release-triage skill You: "Use the release-triage skill on this failing build." the skill loads, then calls UniversalBench: search the web for the error (web search) read the failing file (GitHub) run a fix in the sandbox (code) commit when it passes (validated commit, rollback ready)

Skills run code, so the runtime matters

Anthropic is blunt about this in its own guidance: a skill can direct your AI to run code or call tools in ways that do not match its stated purpose, so you should only run skills you trust. When a skill’s steps run on UniversalBench, three checks are always on. Code is validated and confirmed to load before it is ever committed. Every model call is cost estimated and capped before it runs. No request can reach a private or internal network address. The work gets done without handing a skill the keys to everything.

Connect UniversalBench

Add the MCP server

In Claude Code, add the server with claude mcp add or your MCP config, pointing at your UniversalBench URL.

Keep your skills folder

Your SKILL.md files stay on the filesystem where Claude Code already reads them.

Save secrets in the vault

Store tokens and database URLs once in your secret vault.

Run heavier steps hosted

When a skill needs a hosted sandbox, your database, the web, or a validated commit, it runs through UniversalBench.

# your UniversalBench MCP URL https://universalbench-mcp.penantiaglobal.workers.dev/u/<your-key>

Credentials live in an encrypted secret vault

When a skill’s step needs a credential, a GitHub token, a database URL, an API key, you save it once in your secret vault. You enter the secret name on the left and the secret value on the right, then save. Your AI never sees the value, it is never shown in the chat, and the matching capability unlocks automatically.

UniversalBench secret vault: the secret name on the left, the secret value on the right, then save
Any secret goes in your secret vault once: the secret name on the left, the secret value on the right, then Save.

A skill alone versus a skill with an execution layer

Skill on its ownSkill plus UniversalBench
What it gives the AIInstructions and small scriptsInstructions plus real execution
Run code, query a database, hit the webOnly if the host provides itOne hosted URL covers all of it
Reach your toolsDepends on the surfaceGitHub, Postgres, Slack, Sheets, and more
Safety on the workUp to you to auditValidated code, capped spend, no internal network
CredentialsIn files or promptsEncrypted vault, never shown

Frequently asked questions

How is this different from a plain Claude Code skill?

A plain skill runs on your machine. Adding UniversalBench gives a skill a hosted sandbox plus your database, the web, and validated commits through one MCP URL, useful when you do not want the work running locally or unchecked.

How do I add it to Claude Code?

Add the server with claude mcp add or your MCP config, pointing at your UniversalBench URL, then your skills can call it.

Does it work with other clients?

Yes. The same URL works with Claude, ChatGPT, Gemini, Cursor, and any MCP-compatible AI.

Is it official?

No. Claude Code and skills are Anthropic. UniversalBench is an independent execution layer you connect.

More

Run your Claude Code skills hosted

One URL, first 1,000 calls free, no credit card.

Get your API key →