Google Agent Skills: Skills for Google Cloud Agents
/skills
Google Agent Skills is a GitHub repository of installable Agent Skills that teach an AI coding agent how to work with specific Google Cloud and Google product APIs, from GKE and BigQuery to the Google Ads API. Reach for it if your agent already talks to Google Cloud and you want fewer guessed gcloud flags and stale console steps; skip it if your stack lives outside Google Cloud, since every skill here is Google-product-specific.
Understanding Agent Skills for Google Products
Google Agent Skills is a curated collection of Markdown-based Agent Skills, structured instruction files an AI coding agent reads to learn how to operate a specific Google product, from authenticating to Google Cloud to configuring a GKE cluster or calling the Google Ads API. Each skill lives in its own folder under skills/cloud, skills/ads, or skills/analytics, and you install only the ones you need via npx skills add google/skills.
Available Skills and Covered Technologies
- ✓Getting-started skills for Google Cloud — authentication, a foundation-builder recipe, and an onboarding skill — for standing up a new environment from scratch.
- ✓Multi-product solution skills that chain several Google Cloud services together, like a RAG-for-enterprise-search skill spanning GKE and AlloyDB, or an agentic analytics workflow across Spark and a knowledge catalog.
- ✓AI/ML skills covering the Agent Platform (alerts, endpoints, model registry, tuning, prompt management, RAG engine), BigQuery AI & ML, and the Gemini API, Live API, and Interactions API.
- ✓Infrastructure skills concentrated on GKE — cluster creation, autoscaling, networking, storage, multi-tenancy, upgrades, workload troubleshooting, and TPU/GPU-specific disruption handling.
- ✓Database and analytics skills for AlloyDB, BigQuery (including BigFrames), Bigtable, Cloud SQL, and Spanner, plus a data-lineage summary skill.
- ✓A full set of Well-Architected Framework skills, one for each of the six pillars: Cost Optimization, Operational Excellence, Performance Optimization, Reliability, Security, and Sustainability.
- ✓Advertising and analytics skills outside core Cloud — the Google Ads API, Data Manager API, Google Mobile Ads SDK (banner, interstitial, rewarded formats), and the Google Analytics Admin and Data APIs.
Adding Skills to Your Agent Project
Google Agent Skills installs with one command: npx skills add google/skills. Running it opens a picker so you select the specific skills you want from the repo instead of pulling in the whole collection. The README doesn't document any other install path — there's no separate npm or PyPI package named in it, just the npx command against the GitHub repo itself.
Integrating Skills with Agent Harnesses
Beyond installing individual skills, the repo also bundles Google product plugins — combining Skills and MCP servers — for specific agent harnesses. Claude Code users add the marketplace with claude plugin marketplace add google/skills, then run claude plugin install <plugin>@google-plugins. Codex users add the marketplace with codex plugin marketplace add google/skills, then install a plugin from the /plugins browser. Antigravity CLI users run agy plugin install pointed at a specific plugin path in the repo. The README doesn't document a plugin path for any other agent beyond these three.
Strengths
- ✓Skills are organized by real Google Cloud domains (Infrastructure, Databases and analytics, AI/ML, Advertising, etc.), so picking the right one is fast instead of digging through one giant catalog.
- ✓The npx skills add google/skills installer lets you select just the skills you need instead of pulling the whole repo into your project.
- ✓Plugin bundles ship for three different agent harnesses — Claude Code, Codex, and Antigravity CLI — so the same skill content isn't locked to one vendor's agent.
- ✓Maintained directly under the google GitHub org and licensed Apache-2.0, with skills for genuinely deep, product-specific territory (GKE TPU dynamic-slice monitoring, cross-project Cloud Logging config) beyond basic getting-started material.
- ✓Skills read as structured Markdown per product, which keeps the "how do I configure this GKE cluster" answer scoped to Google's own documented steps rather than the agent's general training knowledge.
Current Development Status and Considerations
- △The README flags the repo as "under active development" — no version numbers or stability guarantees are given for individual skills.
- △Every skill is scoped to a specific Google product; there's no general-purpose or multi-cloud skill in the collection.
- △Infrastructure skills skew heavily toward GKE — teams on Cloud Run, Firebase, or other non-Kubernetes compute get a comparatively thin set (Cloud Run Basics and Firebase Basics are the only entries under "Web and app hosting").
- △Using the packaged plugin bundles, as opposed to picking individual skills via npx skills add, requires an agent harness that already supports the plugin format — Claude Code, Codex, or Antigravity CLI are the three named in the README.
- △The README doesn't document a testing or review process behind individual skill files beyond inviting users to report bugs or inaccuracies through GitHub issues.
Related Google Skill Repositories and Plugins
Frequently Asked Questions
Google Agent Skills are Markdown-based instruction files in the google/skills GitHub repository that teach an AI agent how to work with a specific Google product, from authenticating to Google Cloud to configuring a GKE cluster or calling the Google Ads API.
Google Agent Skills are installed with the command npx skills add google/skills, which lets you pick individual skills from the repo rather than pulling in the entire collection.
Google Agent Skills cover Google Cloud services across categories including GKE, BigQuery, AlloyDB, Bigtable, Cloud SQL, Spanner, Cloud Run, Firebase, Cloud Monitoring, Cloud Logging, and the six Well-Architected Framework pillars, plus Google Ads and Google Analytics APIs outside Cloud proper.
Google Agent Skills accepts contributions through GitHub issues — you can report bugs or inaccuracies in a skill's Markdown file, or file a feature request to suggest a new skill or Google technology to cover, per the README's contributing section.
The README marks the google/skills repository as under active development, so Google Agent Skills should be treated as evolving rather than a fixed, versioned release — check the specific skill file's content before relying on it in production.
Google Agent Skills is released under the Apache-2.0 license, per the repository's license file, so you're free to copy, modify, and distribute the skill files.
The problem it solves
An AI agent asked to wire up a GKE cluster, tune a BigQuery AI/ML model, or diagnose a TPU disruption is working from general training knowledge about Google Cloud, which drifts out of date and doesn't know which of several overlapping GKE networking or autoscaler options your workload actually needs. Google Agent Skills packages that missing context as per-product instruction files — one for gcloud CLI basics, one for GKE cluster autoscaler behavior, one for cross-project Cloud Logging setup — so the agent reads current, specific steps instead of guessing.
Best use cases
- •Bootstrapping a brand-new Google Cloud environment end-to-end with an agent, using the auth, foundation-builder, and onboarding skills together instead of walking through the console yourself.
- •Debugging a specific GKE problem — a stuck JobSet, a TPU VBAR out-of-memory error, a disrupted GPU workload — with a skill written for that exact failure mode rather than generic Kubernetes advice.
- •Running a Well-Architected Framework review against one pillar at a time, such as Cost Optimization or Reliability, from inside an agent conversation instead of a manual audit.
- •Connecting an agent to Google Ads API or Data Manager API workflows — account diagnostics, audience or event ingestion — without re-explaining the setup each session.
- •Standing up a Claude Code or Codex plugin that bundles several of these Google Cloud skills with an MCP server, instead of wiring the integration by hand.
Who should try it — and who should skip
Reach for Google Agent Skills if your agent already operates against Google Cloud or a Google product API and you want it grounded in specific, current steps instead of general training knowledge — pull in just the GKE, BigQuery, or Google Ads skills you need via npx skills add google/skills. Skip it if your team runs outside Google Cloud entirely, or if you need one polished, stable skill rather than a fast-moving collection still under active development.
Related repositories
Still deciding about skills?
One click hands the question to an AI along with this page — see what it says about skills.