Cloudish Skills

Agent Skills for Cloudish, the cloud an AI agent deploys a container to.

Cloudish lets an agent create its own API key with one unauthenticated call,
build any Dockerfile or source folder server-side, and run it as a long-lived
container with an optional persistent volume, paid from prepaid credits the key
can never exceed. No browser, no card, no local Docker.

Skills

The skill is a copy of https://cloudish.ai/skill.md and the docs it links
to (credits, instances, replicas, container isolation, changelog), bundled so
the agent reads them locally. It never fetches instructions at runtime; new
versions ship as new releases of this plugin.

What it runs and sends

The plugin contains instructions only: no MCP server, hooks, scripts, or
binaries. When you ask it to deploy, the agent uses curl and tar to call
the Cloudish API at https://cloudish.ai, and to check that your deployed app
answers at its own URL (https://<name>.cloudish.app/, or your own domain):

• Sends a .tar.gz of the project folder (excluding .git and .env*) to
  Cloudish to be built and hosted, plus the project name, port, how it should
  run (always on, or sleeping when idle), and any env vars or secrets you ask
  it to set.
• Creates a Cloudish API key with an unauthenticated call when the
  project doesn't have one yet. The key is used only to authenticate API calls
  to cloudish.ai, and is never printed to chat.
• Reuses that key on later runs instead of minting another, so the project
  keeps the same Cloudish balance. It's a key the skill created on Cloudish,
  not one of your other credentials. A sub-agent can get its own key, minted
  and funded by the parent agent's key.
• Reports the agent product and model (for example claude-code,
  claude-opus-5-5) when it creates a key, if known, plus skills-repo
  as the source. All three are optional.

Adding credits happens outside Claude: the agent gives you a one-time
www.cloudish.ai link, and any payment is made on that page in your browser.
If your app signs its users in with Cloudish, that sign-in runs between your
deployed app and www.cloudish.ai, not through the agent.

The bundled docs are generated from the ones served at cloudish.ai, with links
between them pointing at the local files.

Your deployed code and data are handled under Cloudish's
privacy policy and
terms.

Install

Claude Code

/plugin marketplace add cloudishai/skills
/plugin install cloudish@cloudishai

Any agent that reads SKILL.md (Claude Code, Codex, Cursor, …) — copy
the skills/cloudish folder into your agent's skills
directory.

Then ask your agent: "Deploy this app to Cloudish."

In the Claude apps and Cowork, the agent runs in a sandbox: when asked,
allow outbound network access to cloudish.ai (the API) and cloudish.app
(where your deployed apps are served). Keys don't carry over between chats
there, so claim yours (the agent can give you a link) to keep your projects
and credits with your account.

Layout

skills/cloudish/SKILL.md          the skill (copy of cloudish.ai/skill.md)
skills/cloudish/*.md              the docs it links to, bundled
.claude-plugin/                   Claude Code plugin + marketplace manifests
plugin.json                       OpenAI / Agent Plugins manifest
assets/                           icon

License

MIT