Google Skills: 186 Free Agent Skills for Google Cloud

Here’s a sentence that would have sounded strange a year ago: the most useful thing Google published for developers this week isn’t a model or a console update — it’s a folder of instructions. The repo is google/skills, and it’s trending at #9 on GitHub today with 532 new stars in a day and 22,640 total. It contains 186 Agent Skills: step-by-step playbooks that teach an AI coding agent how to do Google Cloud work the way Google recommends doing it — setting up projects, tuning Gemini models, building RAG pipelines, debugging GKE clusters, writing BigQuery queries, and dozens more.

The skills themselves are free and open source under Apache 2.0. Installing them takes one command. The catch, and it’s a real one, is that the work they do happens in your own Google Cloud project — and Google Cloud bills by the hour. The skills are free. The infrastructure they spin up is not.

What an Agent Skill actually is

An Agent Skill is a folder containing a SKILL.md file: metadata (a name and a description) plus step-by-step instructions, optionally bundled with scripts, reference docs and templates. The format was originally developed by Anthropic and released as an open standard. Agents load skills through progressive disclosure — only the names and descriptions load at startup, the full instructions load when a task matches, and the agent follows them from there. Installing the whole 186-skill catalog costs almost no context until a skill is actually used, which is why a catalog this big is practical at all.

That design is what makes this repo more than a documentation dump. A skill for tuning a Vertex AI model doesn’t just link to the docs; it walks the agent through the workflow — which commands to run, which parameters matter, which checks to do before spending money. The difference between reading documentation and following a skill is the difference between knowing how something works and having someone experienced standing over your shoulder while you do it.

One command to install

npx skills add google/skills

That’s the whole install. The installer prompts you to pick specific skills or all of them, which target agents to configure, and whether the scope is a single project or global. To update later:

npx skills update

The only prerequisite is that npx is available — install Node.js if it isn’t, since npm and npx come with it. There are also plugin routes for the three big agent harnesses: Claude Code (claude plugin marketplace add google/skills, then install the plugin), Codex (same pattern through its plugin browser), and Antigravity CLI (agy plugin install with the plugin path). The README doesn’t enumerate the full list of agents the installer supports, so if your agent isn’t one of those three, run the installer and see what it offers rather than assuming.

What’s inside: 186 skills, and a lot of GKE

The catalog is deep rather than broad — it covers Google’s world, and only Google’s world. The breakdown, counted from the repo’s own index, runs like this:

• Getting started (3 skills): onboarding to Google Cloud, authenticating and authorizing with IAM, and a Foundation Builder recipe. The onboarding skill is the one to read first: it requires a personal Google account and a valid payment method for identity verification and to activate the $300 free-trial credit, and it walks through gcloud auth login, project creation and linking a billing account. Deploying anything requires billing enabled.
• AI/ML (28 skills): Gemini and the Vertex AI Agent Platform, including RAG Engine management, an Eval Flywheel, model tuning and tuning management, Model Garden deployment, Model Registry, endpoint management and prompt management; Genkit in Dart, Go, JavaScript and Python; LiveAPI; BigQuery AI and ML; and an API Gateway remote MCP server.
• Infrastructure (44 skills): heavily GKE-weighted — around 25 of them cover cluster creation, autoscaling, networking, storage, upgrades, TPU troubleshooting, cost analysis and reliability — plus Cloud Storage, Compute Engine reservations, Filestore, load balancers and networking observability.
• Databases and analytics (19): BigQuery (basics, optimization, troubleshooting, UDFs, observability, migration and a SQL translator, BigFrames), AlloyDB, Bigtable, Cloud SQL, Spanner, Managed Airflow and data lineage.
• Security and identity (23): IAM in several flavours, Google SecOps (case management, detection engineering, incident and entity investigation, alert triage, threat hunting), Security Command Center, Cloud Armor, Assured Workloads, Sign In With Google and DPoP identity security.
• Management tools (19): Cloud Monitoring (chart generation, PromQL, alert configuration, notification channels, an SLO wizard) and Cloud Logging, plus Cloud Trace and GKE observability and cost.
• Well-Architected Framework (6): one skill per pillar — cost optimization, operational excellence, performance, reliability, security and sustainability.

Seven more skill packs live in separate repos — Advanced Google Cloud Storage, Agent Development Kit, Android, Dart, Firestore, Flutter and Google Maps Platform — so the 186 in this repo is the core, not the ceiling.

The honest caveat: the skills are free, the cloud is not

This is the part the star count doesn’t tell you. A skill is text. It tells an agent how Google recommends doing something, and then the agent executes real, billable commands against your own project — creating projects, linking billing, deploying infrastructure. A misread step can spin up chargeable resources, and the skills automate exactly the parts of that process (enabling service APIs, creating resources) where costs start accumulating.

To Google’s credit, the onboarding skill builds in guardrails: pre-execution audits, mandatory consent before project or billing changes, and spend-control chaining. Read the skill before you run it, and read it again before you let an agent execute the expensive steps. The $300 first-time free trial and the Free Tier give you room to experiment, but the trial is a one-time cushion, not a permanent shield.

Two more honest notes. First, coverage is Google-only and vendor-written: the content reflects Google’s recommended way of doing things, and it can go stale as services change. “Google’s own playbook” is a fair framing; an independent benchmark it is not. Second, roughly a quarter of the catalog is GKE. If your work lives on Google Cloud, that’s a gift. If it doesn’t, most of this repo won’t apply to you — and that’s fine, because the install lets you pick individual skills instead of all 186.

Who should install this

If you already use Claude Code or Codex and you already work in Google Cloud, install it today — the onboarding and IAM skills alone will save you the kind of fumbling that eats an afternoon. If you’re learning Google Cloud, it’s arguably the best structured path into the platform that currently exists, because it teaches by doing real tasks instead of presenting reference pages. If you don’t use Google Cloud at all, skip it; a playbook for a platform you don’t run is just a reading list.

And if you’re the kind of person who reads trending repos to find what working developers are actually adopting, the signal here is worth noting: 22,640 stars since March, 546 commits, and a catalog that grew to 186 skills in seven months. That’s not a demo. That’s a team shipping a new kind of documentation — one written for the agent sitting next to you, not the human reading the manual.

For more on the AI-agent wave, see our Meta Muse Spark Open release, Anthropic’s Claude Duet agent, and today’s Meta Muse voice rollout.

Also today: Xbox Cloud Gaming Phone Controller: Play on TV Now and Meta Muse Voice: Talk to Your AI With a Face.

Related: Astryx: Meta’s Free Agent-Ready Design System

Another agent skill worth a look: shadcn/improve audits your codebase with your best model, then writes plans for cheaper models to execute.