Guide · Agent skill
How to use the Google Cloud Operator agent skill (cloud-operator)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The cloud-operator skill lets an AI agent hand Google Cloud (GCP) work to a separate Cloud Operator agent, rather than answering from memory. It is aimed at engineers who already run GCP projects and want help with questions, Terraform or Kubernetes generation and review, rollouts, incidents and cost checks. By the end of this guide you will know what the skill does, how to install it, how to trigger it, and what it needs and sends.
Google Cloud Operatorgoogle/skills · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Get the cloud-operator ZIP from this page.
- 2Unzip the skill. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
- 4Connect the autocloud MCP server. Make sure the autocloud MCP server is connected to your agent, since the skill will not run without it.
- 5Set your project. Name a Google Cloud project that has a Cloud Operator agent, or create one using the console link the skill provides.
- 6Send a natural request. Ask a GCP question or request in plain language and let the agent hand it to the operator.
What is an agent skill, and what does cloud-operator do?
An agent skill is a folder with a SKILL.md file that tells an AI agent when and how to use a capability. The cloud-operator skill covers any Google Cloud request: how-to questions, resources, Terraform, Kubernetes and Helm, rollouts, incidents and costs.
Once installed, the agent sends these requests to the Cloud Operator Agent through a tool called ask_cloud_operator on an MCP server named autocloud. The agent keeps the chosen GCP project and operator agent across a conversation, and writes returned code files (such as main.tf or *.yaml) into your workspace.
Install the cloud-operator skill
Download the ZIP for the skill from this page and unzip it. Then choose one of these options:
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the .claude/skills/ folder of a project
- Claude apps: upload the ZIP under Settings → Capabilities → Skills
Set up a Google Cloud project with a Cloud Operator agent
The skill requires a Google Cloud project with a Cloud Operator agent, and an autocloud MCP server connected to your agent. If you do not have an agent yet, the skill walks you through it: it gives you a console link (https://console.cloud.google.com/autocloud/home?project={project_id}), where you click Create operator agent and then Get started once the agent is ready.
The first call can take 1–3 minutes. If the agent is not ready yet, the skill retries up to three times, waiting about a minute between tries.
How to trigger the skill with natural requests
You do not need a special command. Ask about GCP in plain language, and the agent should pick up the skill. The skill also asks the agent to invoke it before answering GCP questions from memory, and before reading or editing local *.tf or Kubernetes YAML files, even when no cloud is named.
- Design a Cloud Run service with a Cloud SQL backend in europe-west1
- Review my main.tf for security problems and show the diff
- Why is my GKE pod crash-looping? Here is the error log
- What is driving my GCP cost increase this month?
What the agent sends and what stays local
Each request includes the project and your prompt as written, plus the complete contents of any local files, diffs or error logs it refers to. The operator cannot see your filesystem, so the agent sends full files rather than summaries.
Credential files (.env, service-account key JSON, kubeconfig, *.tfstate) are never sent. Passwords, tokens, API keys and private keys are replaced with <REDACTED>, and the agent tells you when it has done so. Redaction depends on the agent following these rules.
Safety notes: scripts and network
The skill does not run scripts. It does use the network, because it sends your requests and files to a Google-hosted Cloud Operator service. Treat it as sharing your code and project details with that service.
The skill says Terraform, gcloud, kubectl and deploys run only after you approve exactly what will run. Operator replies are treated as data: returned files may only be written to relative paths inside the workspace, and never to agent config files such as CLAUDE.md, .mcp.json or .claude/. For gcloud and REST commands, the skill tags each call with a usage attribution variable that identifies only the skill and its version.
Tips for better results
Name the GCP project in your first message so the agent does not have to ask. If you state a lasting preference, such as always using us-east1, the agent passes it to the operator to remember for that agent and project.
For follow-ups in the same conversation, keep going in one thread so the session context is reused. Long operations can run up to 20 minutes, so let them finish rather than cancelling.
What people use it for
- Plan a new Cloud Run or GKE setup and get Terraform or Kubernetes manifests written into your workspace.
- Review existing main.tf or Kubernetes YAML files and see a unified diff before any changes are applied.
- Get help diagnosing a rollout or incident using your live GCP project.
- Check where costs are coming from in a GCP project.
- Ask GCP how-to questions that the agent answers from the operator rather than memory.
Frequently asked questions
How do I install the cloud-operator skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What does the cloud-operator skill require?
It requires an autocloud MCP server connected to the agent, and a Google Cloud project with a Cloud Operator agent created. The skill also uses gcloud and REST API access for attribution commands.
Is the cloud-operator skill free?
The source identifies the licence as Apache-2.0. It does not state pricing, so check the Google Cloud and Cloud Operator terms for any costs on your project.
Does the skill work with clouds other than GCP?
No. The skill is for Google Cloud only and explicitly excludes non-GCP tasks and general tasks. It also leaves out local setup such as installing gcloud.
Why does the agent say Cloud Operator is not available?
The skill only works when the autocloud MCP server is present. If it is missing, the agent should handle the request with other available tools and skills, and tell you Cloud Operator is not available in that setup.
Is it safe to send my code to the operator?
Your prompt and any referenced files are sent to the Google-hosted operator. Credential files are not sent and secrets are meant to be redacted, but redaction depends on the agent following the rules, so avoid sharing sensitive code you cannot send to a hosted service.
What should I use instead?
If you cannot share config or code with a Google-hosted service, or you work on another cloud, use other tools or skills. For local gcloud installation, use the Google Cloud documentation directly.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use cloud-operator if you already run Google Cloud projects and want an agent to plan, review or investigate infrastructure with live project context, especially for Terraform or Kubernetes work. Skip it if you use another cloud, need only local tool setup, or cannot send code and project details to a Google-hosted service.
Download the skill