What Are Google Agent Skills? A Freelancer's Guide to Google's Official Skills Repository

In early November 2026, Google published an official, open-source repository of Agent Skills — instruction packages that teach an AI coding agent to work with Google Cloud, Firebase, BigQuery, the Gemini API and more, the way Google recommends. The repository, google/skills on GitHub, had already gathered around 23,000 stars as of 6–7 Nov 2026. For freelancers it matters: clients are asking for Gemini-powered apps and Google Cloud setups, and AI-assisted delivery is the faster route.

What Are Google Agent Skills?

An Agent Skill is a folder of instructions — usually built around a file called SKILL.md — that tells an AI coding agent how to perform a specific task well. Instead of guessing how to call an API or configure a service, the agent gets step-by-step procedures, the current SDK patterns, and the pitfalls to avoid.

Agent Skills in one paragraph

The key idea is progressive disclosure. A skill's SKILL.md is a short summary the agent loads first, describing what the skill does and when to use it; supporting files — scripts, reference documents and examples — load only if needed, keeping the agent's context window free. The format began as an open standard from Anthropic in December 2025, now supported by more than 26 platforms.

Why Google built an official repository

Before this repository, an AI agent working with Google Cloud had to rely on training data that is often months out of date. That is how agents end up using deprecated SDKs or inventing API parameters that do not exist. Google's AI Studio docs note the Gemini skills give agents current SDK patterns, avoiding deprecated SDKs. The skills are open source under the Apache 2.0 licence, and anyone can read exactly what the agent is being told — so you can inspect a skill before letting it touch a client's project.

What's Inside google/skills? The 13 Launch Skills

Product skills

The seven product skills cover the Google services freelancers are most likely to be asked about:

• Gemini API — building apps with the Gemini API, including the Live API and Gemini Omni Flash.
• Firebase — projects, authentication, databases and hosting.
• BigQuery — queries and analytics on Google's warehouse.
• Cloud Run — deploying containerised services that scale to zero and back up on demand.
• Cloud SQL — managed relational databases and day-to-day operations.
• AlloyDB — Google's fully managed PostgreSQL-compatible database service.
• GKE (Google Kubernetes Engine) — containerised workloads on Kubernetes.

Well-Architected pillar skills

The three pillar skills are about doing cloud work properly: Security (identity, access, secrets, network controls), Reliability (uptime and failure handling), and Cost Optimisation (finding wasted spend and right-sizing resources). These are quietly the most valuable for freelancers: clients rarely ask for a "reliability review" by name, but most worry about security and cost, and a skill that guides your agent through a structured review turns that worry into a service you can sell.

Recipe skills

The three recipe skills cover repeatable workflows across products: Authentication, Onboarding and Network Observability — sign-in done right, new projects set up right, and network problems traced when things go slow.

How to Install Google Agent Skills (Step by Step)

One-command install

Installing the whole repository takes a single command, run in your project directory:

npx skills add google/skills

Which agents they work with

Google lists plugins for Claude Code, Codex and the Antigravity CLI, and the skills also work in Cursor, Gemini CLI and VS Code. If you already use one of these agents, add the skills to the agent you already have.

If you work in Cursor, get proper access to the tool itself before installing skills into it — Cursor Pro on ProAccess.me gives you the full Cursor experience these skills are designed for. Likewise, the Gemini API skill needs Gemini access — Gemini AI Pro on ProAccess.me is the straightforward way to get it for client work.

Picking only the skills you need

You do not have to install all 13 skills — and you should not. Each skill is context your agent has to carry, and one you never use is just noise. Start with the two or three that match your work: Gemini API for AI features, Firebase for apps, cost-optimisation for cloud reviews.

How Freelancers Can Turn Agent Skills Into Client Work

Agent Skills are free instructions. The billable part is what you do with them: judgement, client communication, testing, and taking responsibility for the result.

Service idea 1 — Google Cloud setup and migration

Small businesses moving to Google Cloud need the same thing: a project set up correctly, authentication wired up safely, and their app or data moved across without breaking. The onboarding and authentication recipe skills guide your agent through exactly that sequence. Package it as a fixed-scope offer — "Google Cloud setup in one week, including auth and a handover document."

Service idea 2 — Gemini API apps for clients

The Gemini API skill is the most directly commercial of the 13. Clients want chatbots, document summarizers and content assistants built on a model they have heard of. With the skill guiding your agent to the current Gemini SDK patterns, you can prototype a working demo in a first client call. Quote the build as a fixed-price project; API usage itself is billed by Google — the skill makes building faster, not free.

Service idea 3 — Cost-optimisation and security reviews

Many small companies run Google Cloud setups configured once and never reviewed. The cost-optimisation and security pillar skills give your agent a structured checklist to audit one: oversized instances, forgotten resources, open permissions, missing alerts. Offer it as a one-off review with a written report, priced as a flat fee. It is a low-risk first engagement, and it often leads to larger follow-up work.

How to price and package it

Fixed-scope starter offers work best. Pick one service, define exactly what is included, set one price, and put a clear boundary around it — for example, "one Firebase project, auth included, delivered in five working days." Clients buy certainty, and a fixed scope protects you from open-ended work.

One practical tip: if your client work involves repeated, multi-step workflows — intake forms, status updates, handover checklists — an automation tool can carry the routine parts. n8n Starter on ProAccess.me is a good fit for wiring agent-driven work into an automated delivery pipeline.

The Limits and Risks to Respect

Skills are instructions, not permissions — cloud spend and access still apply

A skill tells your agent how to do something. It does not grant access, waive quotas, or make cloud usage free: every BigQuery query, Cloud Run deployment and Gemini API call is still billed to the project it runs in. Before running a skill on a client's project, confirm the project, billing account and spending limits.

Check the skill before you trust it

The repository is new and actively changing. One review flags documentation gaps after a restructure (Issue #33) and an auth problem with the Developer Knowledge MCP server (Issue #223). Separately, a January 2026 analysis found that roughly 1 in 4 public skills contain at least one vulnerability — a figure covering public skills in general, not Google's repo specifically, but still a reason to read a skill's instructions before running it, especially one that asks for cloud permissions.

Never run a skill against a client's live project untested

This is the single most important rule in this guide: test every skill in a safe project first. Create a throwaway Google Cloud project, run the skill there, check what it creates, and only then use it on client work. Your client is paying for your caution as much as for your speed.

Google Agent Skills vs Other Agent Skill Sources

Here is how it compares with the other common sources:

Source	What it covers	Who maintains it	Best for
google/skills (official)	Google Cloud, Firebase, Gemini API, GKE	Google	Client work on Google products
Community skill registries	Thousands of skills for many tools	Community contributors	Niche tools Google does not cover
Your own custom skills	Your exact workflows and client standards	You	Repeatable work you do every week

The sensible approach is to combine them: Google's official skills where they exist, community skills for tools Google does not cover, and your own skills for the workflows you repeat for clients. Your own skills are the long-term asset — they capture how you work, and they compound with every project.

FAQ

Are Google Agent Skills free?

Yes — the repository is free and open source under the Apache 2.0 licence. The cloud usage the skills drive is not: Google Cloud services and Gemini API calls are billed normally to the project they run in.

Do I need a Google Cloud account to use them?

You need a Google Cloud account with a project and billing set up before a skill can do anything useful. You can install and read the skills without one, but running the BigQuery, Cloud Run or Gemini API skills needs a configured project.

Which coding agents support Agent Skills?

The format is an open standard supported by more than 26 platforms, including Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor and VS Code. Google lists plugins for Claude Code, Codex and the Antigravity CLI.

Can I use them with Cursor / Claude Code?

Yes. Both support the Agent Skills format. Install them with npx skills add google/skills in your project, and the agent will pick up the relevant skill when the task calls for it.

Are Agent Skills the same as MCP servers?

No, although they are often used together. An MCP server gives an agent a live connection to a tool or data source; an Agent Skill gives it instructions on how to do a task well. A skill can tell the agent to use an MCP server, but the skill itself is documentation, not a live connection.

Conclusion

Google Agent Skills are a small, practical gift to anyone who builds on Google's platform: official, open-source instructions that stop your coding agent from guessing. The 13 skills at launch cover the products freelancers are most often asked about, plus the security, reliability and cost reviews that turn one-off builds into ongoing client relationships.

The sensible first step this week is small. Pick one skill that matches work you already do, install it with npx skills add google/skills, and try it in a test project — not a client's live environment. Build one small thing with it, read what it did, and decide whether it earns a place in your workflow. If it does, you have the start of a service you can offer clients.