google/skills is Google official open source collection of Agent Skills aimed at Google products and technology, with a focus on Google Cloud. It packages GCP onboarding, Vertex AI solutions, and operations know how into skills your agent can call directly, so cloud engineers and AI app developers spend less time digging through docs and stitching best practices together. This repo is the latest big tech move, after NVIDIA and Microsoft, to hand their cloud best practices straight to agents, and it signals that agent skills have moved from personal projects to official infrastructure.

What it does

google/skills follows the open Agent Skills format at agentskills.io, with each skill living in its own directory and a SKILL.md explaining it. The content splits into three blocks. GCP onboarding covers IAM authentication and authorization, Foundation Builder, and onboarding guidance. Multi product solutions include solution architecture workflows, multi agent security, cross cloud agentic analytics, boundaryless data lakehouse, building and deploying AI agents, enterprise RAG search, and strict private tier n tier serverless web apps. AI and ML platform operations cover access auditing, alerting setup, endpoint management, evaluation flywheel, GenAI inference, model garden deployment, model registry, model fine tuning, and prompt management. It is essentially a GCP cookbook broken into order able skills, each scoped to a concrete task.

What problem it solves

Cloud engineers and AI app developers working on GCP often face a wall of documentation and have to assemble best practices by hand. google/skills distills that experience into agent callable skills, and a single npx skills add google/skills bakes knowledge like how to go to cloud, how to do multi agent security, and how to use Vertex AI into place. That lowers the cost of starting from zero on GCP and keeps team knowledge from walking out the door when people leave. For newcomers it acts like an always available senior GCP colleague who can save a lot of wrong turns, and for veterans it shortens the gap between a question and a verified operational answer.

Language and architecture

The main language is Python, and skills follow the open Agent Skills format with topics tagged google, googlecloud, and skills. Install and update go through the skills CLI. After npx skills add google/skills you choose specific skills or the full set, pick a target agent such as Claude Code, Cursor, or Codex, and choose a scope such as project level or global. Later npx skills update upgrades and can clean up stale skills. The repo was created in March 2026 and is actively pushed, continually expanding GCP and AI and ML skills roughly in step with Google Cloud capability releases, which means the skills track the platform instead of going stale.

License

The project is Apache 2.0, commercially friendly, and safe to plug into internal toolchains, or fork and trim into a team only subset that reflects your own conventions.

After NVIDIA, Microsoft, and scenario labs, Google official Agent Skills entered the field, handing GCP and Vertex AI best practices directly to agents and riding the agent skills wave of 2026. It gained about 534 stars on its launch day. It belongs to the same wave of official skill repos as What Is cloudflare/security-audit-skill and How to Use It, but unlike What Is github/spec-kit and How to Use It, which is a development workflow tool, google/skills leans toward cloud and AI operations and is tied to a specific cloud vendor.

Where to find it

GitHub repo https://github.com/google/skills , skills directory https://skills.sh/google/skills .

Getting started

Run npx skills add google/skills into a project level scope, then pick one or two scenarios you are actively working on, such as onboarding or Vertex AI inference, and let the agent call them to see if the steps fit your environment. Once it proves useful, consider a full install and build a habit of npx skills update so you are not using outdated cloud best practices, and pin a known good version if you depend on it in production.

FAQ

How is this different from just asking Gemini about GCP? The difference is that curated, operative knowledge is baked in as a skill, so agent calls are more predictable and closer to the officially recommended path than an open ended chat.

Do I have to use the skills CLI? It is the recommended path for install and upgrade right now, since it auto adapts to different agents directory layouts and versions, and it handles cleanup of stale skills.

Is it locked to Google ecosystem? Yes, the content centers on GCP and Vertex AI, which is great for teams already on or moving to Google Cloud, and less useful for purely local or other cloud work.

Can I contribute skills back? The repo is open source, so you can propose skill additions or fixes through the normal pull request flow, which also helps the skills stay current.