How to use the Cloud Run Basics agent skill

What the cloud-run-basics skill makes the agent do

Cloud Run has three resource types. Services respond to HTTP requests at a stable endpoint and autoscale their stateless instances. Jobs run tasks manually or on a schedule until they finish. Worker pools handle always-on background work such as pull-based consumers of Kafka, Pub/Sub or RabbitMQ.

With the skill installed, the agent can help you choose which resource type fits your workload, enable the Google Cloud APIs it needs, grant the required IAM roles, and deploy either a container image or source code using gcloud commands.

Install the skill: download the ZIP and unzip it

Download the ZIP for this skill from this page and unzip it. For a personal install, place the skill folder in ~/.claude/skills/. To scope it to one project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Single project: .claude/skills/ inside the project
• Claude apps: Settings → Capabilities → Skills (upload)

Set up your Google Cloud project before you start

The skill needs the Google Cloud SDK (gcloud CLI), a Google Cloud project with billing enabled, and permission to enable APIs and grant IAM roles. It cannot create a project or billing account for you.

Enable the Cloud Run Admin and Cloud Build APIs with the command below. The skill's guidance uses these APIs for deployment.

• Google Cloud SDK (gcloud CLI) installed and authenticated
• A project with billing enabled
• Roles: Cloud Run Admin, Cloud Run Source Developer, Service Account User, Logs Viewer
• Cloud Build service account granted Cloud Run Builder (roles/run.builder) for source builds

Trigger the skill with natural requests

You do not need special syntax. Describe the outcome you want in plain language, and mention Cloud Run or Google Cloud so the agent knows which platform you mean. The skill's description covers deploying HTTP-serving applications (services), event-triggered or scheduled tasks (jobs), and pull-based background processing (worker pools).

Good requests name the workload and the goal, for example: deploy this Flask app to Cloud Run, or set up a nightly batch job on Cloud Run that runs my container.

• Deploy my Node API from this folder to Cloud Run.
• Which Cloud Run resource type should I use for a Pub/Sub consumer?
• Schedule a daily job on Cloud Run using my image.
• Show me the IAM roles my deploy account needs for Cloud Run.

Deploy a container image or source code

To deploy a container image, the skill uses gcloud run deploy with an image URL and a region. Service names must be 49 characters or less and unique per region and project. Artifact Registry is the recommended image source; the repository must already exist.

To deploy from source with a Dockerfile or buildpacks, the skill uses --source . so Cloud Build builds the image in the cloud, without Docker needed on your machine. Any deployed code must listen on 0.0.0.0 and use the injected $PORT variable, defaulting to 8080, or it will crash on boot.

• Container image: use an Artifact Registry URL in the form LOCATION-docker.pkg.dev/PROJECT_ID/REPO_NAME/PATH:TAG
• Source: build with Cloud Build via buildpacks or a Dockerfile
• Listen on 0.0.0.0 and read $PORT, not 127.0.0.1 and a fixed port

Safety, network use and public access

This skill does not run scripts, but it does use the network, because its commands call Google Cloud APIs. Before any gcloud, REST API or MCP command, the skill's attribution rules require the agent to read the usage attribution guidance and tag commands with the skill identifier. Those tags are set inline on each command, not saved with gcloud config set.

The example deploy command includes --allow-unauthenticated, which makes the service publicly reachable. Review that flag before running it. If your organization restricts unauthenticated invocations, the skill's guidance points to testing private services instead.

• No scripts are run by this skill
• Network is used for Google Cloud API calls
• Check --allow-unauthenticated before deploying anything private
• Projects under a domain restriction policy may need authenticated access

Tips for getting reliable results

Give the agent your project ID, region and whether the code is a container or source. Those three details decide most commands. Ask the agent to check the required roles first if a deploy fails with a permission error.

Use Artifact Registry for images you control. Docker Hub images are cached for up to one hour, so Google recommends Artifact Registry for higher availability.

• Name your project ID, region (the examples use us-central1) and image or source path
• Ask for a review of the command before it runs
• Use the list command to confirm a deploy: gcloud run services list