Google's Official Agent Skills for Working with Google Cloud

When you ask an AI coding agent to write Google Cloud code, you usually get plausible-looking Terraform and SDK calls that are a release or two behind. Google's official google/skills repository is the maintainer-shipped answer to that problem: a curated bundle of 70+ agent skills that cover Gemini, Agent Platform, BigQuery, GKE, Cloud Run, the Well-Architected Framework, and more, installable on the agents developers already use.

Why This Skill Matters

Skill collections from the platform vendor carry a specific weight: they reflect the current APIs, the current product boundaries, and the current recommended patterns. Third-party collections are useful for breadth, but for accuracy on Google Cloud, this is the source.

The collection spans the things you actually build on GCP:

• AI/ML: Gemini API, Agent Platform, BigQuery AI, Model Garden.
• Data: BigQuery, AlloyDB, Spanner, Airflow, the data lakehouse reference architecture.
• Infrastructure: GKE, networking, Terraform for Google Cloud.
• Serverless & app hosting: Cloud Run, Firebase.
• Well-Architected Framework: all six pillars as separate skills.
• Identity & security: GKE security, SecOps, IAM guidance.
• Operations: Cloud Monitoring, Cloud Logging, cost optimization.

For anything Google-Cloud-shaped, this is the collection to check before you prompt.

Installation

npx skills add google/skills

Then interactively select which skills to install. You'll get a ~/.agents/skills/ (or host-equivalent) directory with the chosen skills ready to load.

Real Workflow: Migrate an AI Workload from AI Studio to Agent Platform

A realistic scenario, not "hello world": you prototyped an agent in AI Studio, and now you need to move it to Google Cloud's Agent Platform with proper endpoint management.

The workflow:

1. Activate agent-platform-migrate-from-ai-studio to learn the migration path.
2. Activate gcloud CLI skills for auth and project setup.
3. Activate agent-platform-endpoint-management to deploy the model on Model Garden.
4. Activate cloud-monitoring to wire up observability.

Prompt your agent:

Migrate this AI Studio prototype to Google Cloud's Agent Platform.
Authenticate via gcloud, deploy the model on Model Garden with
endpoint management, and wire up Cloud Monitoring for the new endpoint.
Use the matching skills: agent-platform-migrate-from-ai-studio,
gcloud, agent-platform-endpoint-management, and cloud-monitoring.

Expected output: a step-by-step migration plan (project setup, IAM, endpoint creation, monitoring), concrete gcloud invocations, and a checklist of what's left to verify. Because each skill encodes the current Google Cloud patterns, the agent reasons over the right shape — not the AI-Studio shape.

Real Workflow: Apply the Well-Architected Framework to an Existing Service

Another realistic scenario: you have a Cloud Run service in production and want to apply Google's Well-Architected Framework without re-architecting everything.

The workflow:

1. Activate each of the six Well-Architected pillars as separate skills.
2. Have the agent audit your service against each pillar.
3. Synthesize findings into a prioritized remediation list.

Prompt your agent:

Audit this Cloud Run service against the six Well-Architected pillars
(cost, reliability, security, operational excellence, performance,
sustainability). Use the matching skills for each pillar. Produce
a prioritized remediation list with concrete actions.

Expected output: a structured audit with one section per pillar, each containing findings (current state) and actions (what to change). The output is reviewable by an SRE or a security lead because every recommendation cites the pillar it comes from.

Tips

• Install skills selectively. Don't install all 70+ at once; pick the categories your project actually touches. Agent context windows are finite.
• Pair with gcloud for auth. Most of these skills assume gcloud CLI is set up. Get that working first; the rest builds on it.
• Re-pull periodically. The repo banner notes active development — skills can change with new product releases. Pin a version if you need stability.
• Use Well-Architected skills before launch, not after. They work best as a design aid; retrofitting them is much more expensive.

When Not to Use This

• You're not on Google Cloud. These skills are explicitly scoped to Google products. For AWS, Azure, or multi-cloud guidance, use a different collection.
• You need a production SLA on the skills themselves. Help is "via GitHub Issue Tracker" — community-only, no SLA. Plan around that for production-critical workflows.
• You want a turnkey agent runtime. Skills are instructions, not a runtime. You'll combine them with your agent host (Claude Code, Codex, Antigravity).
• You need immutable, audited guidance. The repo is under active development, and breaking changes are possible. If your pipeline requires immutable skill versions, mirror a specific tag.

See the leaderboard for more skills.