google/skills: Google’s official Agent Skills repo for Cloud, Gemini, and recipes

Google’s google/skills repository is a vendor-maintained drop-in for teams that want coding agents to follow Google-flavored playbooks: Gemini, Google Cloud primitives, and Well-Architected guardrails—without rewriting the same onboarding docs in every repo.

This post is a field summary: what’s in the box, how installation works, license posture, and how it sits next to other “skills” terminology across Google’s product surface.

Primary source: github.com/google/skills

TL;DR

Why this matters for builders

Most enterprises already standardize on Google Cloud for data, compute, and identity. The hard part in agent-assisted development is not “can the model write Terraform-ish YAML”—it is whether procedures match internal best practice (networking, security baselines, cost patterns).

A skills monorepo from the vendor compresses that work into reusable agent context: onboarding recipes, service “basics,” and framework-aligned checklists that stay closer to what Google documents than ad-hoc chat instructions.

Skills catalog (as listed upstream)

The README enumerates these packages (names may evolve—check the repo):

• Gemini API in Agent Platform
• AlloyDB Basics
• BigQuery Basics
• Cloud Run Basics
• Cloud SQL Basics
• Firebase Basics
• Kubernetes Engine (GKE) Basics
• Recipe: Onboarding to Google Cloud
• Recipe: Authenticating to Google Cloud
• Recipe: Google Cloud Network Observability
• Google Cloud Well-Architected Framework: Security
• Google Cloud Well-Architected Framework: Reliability
• Google Cloud Well-Architected Framework: Cost Optimization

That mix is deliberately breadth-first: core managed services plus cross-cutting “how we expect you to operate GCP” material.

Installation and operational notes

Upstream documents installation through the skills CLI:

npx skills add google/skills

Practical guidance:

1. Version skills like code—vendor guidance changes with product launches and doc revisions.
2. Combine with MCP where needed—skills teach workflows; live API calls often still belong in MCP servers or your own tooling. See our MCP primer.
3. Review before you trust—treat third-party and vendor skills as configuration you run, not magic. Our agent skills security note still applies to organizational policy.

“Skills” overload: Google has two different meanings

Google also ships Skills in Chrome—saved Gemini prompts for browser workflows. That is not the same artifact as google/skills on GitHub.

We compared the Chrome feature in Skills in Chrome—worth reading if your team is confused by shared naming.

How this fits the broader agent-skills ecosystem

For the cross-tool mental model (metadata, progressive disclosure, MCP pairing), start with What are agent skills?.

For discovery beyond Google:

• Browse skills — community-ranked installs and publisher pages
• MCP servers — when you need live Google APIs in the loop

Bottom line

google/skills is Google’s open, Apache-licensed contribution to the agent skills pattern: packaged operational knowledge for Gemini and Google Cloud that installs with npx skills add google/skills. It complements—not replaces—your org’s security review, MCP integrations, and internal standards.

If you standardize agents across vendors, keep vendor skills (Google) and community skills (explainx.ai registry) in separate namespaces so updates and trust boundaries stay clear.

Repository stats and the exact skill list can change; always verify on GitHub. explainx.ai is not affiliated with Google.