Google ships skills library for agents

Google has started packaging cloud product knowledge into installable “skills” for AI agents, giving developers a public repository they can add to coding assistants and agent runtimes with a single command. The repository, announced April 22 at Google Cloud Next 2026, is positioned as an official source of condensed guidance for products including BigQuery, Cloud Run, Firebase, the Gemini API and Google Kubernetes Engine. (github.com) Google says the goal is to give agents current, product-specific expertise without loading large volumes of documentation into prompts. (cloud.google.com) The release puts Google into a fast-growing layer of the agent stack: reusable connectors and instruction bundles that sit between raw model access and real-world cloud tasks. The public GitHub repository shows 13 skills at launch and is licensed under Apache 2.0, allowing developers to copy, modify and distribute them. ### What exactly did Google ship? Google’s April 22 announcement introduced an “official Agent Skills repository” hosted at `google/skills` on GitHub. The company said the repository launched with 13 skills focused on Google Cloud technologies and workflows. (cloud.google.com) Those include product skills for AlloyDB, BigQuery, Cloud Run, Cloud SQL, Firebase, the Gemini API and GKE, along with recipe and well-architected skills covering onboarding, authentication, network observability, security, reliability and cost optimization. (github.com) GitHub’s README shows the install flow in one line: `npx skills add google/skills`. The repository says users can then choose specific skills from the repo to install. ### Why is Google packaging this as “skills” instead of more docs? Google’s developer blog said traditional approaches such as connecting an agent to a documentation source can create “context bloat,” where too much material is loaded into the model context window, increasing token use and reducing clarity. The company described skills as a compact, open format for giving agents new capabilities and expertise, with content written in Markdown and supplemented by reference files and code snippets. (cloud.google.com) (github.com) Google’s developers blog and ADK documentation describe the same design pattern as progressive disclosure: agents first load lightweight metadata, then fetch fuller instructions and resources only when needed. In a separate April 1 developer guide, Google said this structure can reduce baseline context usage by loading only a skill menu at startup rather than large monolithic prompts. ### Which agent tools can use these skills? Google’s announcement said the repository can be installed into “agents of choice,” naming Antigravity, Gemini CLI and third-party agents. (cloud.google.com) The user-provided briefing also said the skills were built on Anthropic’s open format and aimed at builders using Claude Code, Cursor, Codex and Gemini CLI; that broader compatibility claim could not be independently confirmed from Google’s own announcement pages reviewed here. A Google codelab updated April 10 showed official Firebase agent skills working with Gemini CLI, including instructions for installing and querying those skills locally. (developers.googleblog.com) ### How does this fit into Google’s wider agent push? Google Cloud Next 2026 was framed by the company as a conference about the “Agentic Enterprise,” with Google rolling out a broader stack for building, governing and deploying agents. The skills repository sits inside that push as a developer-facing component, aimed at making agents more useful for specific Google Cloud tasks. (cloud.google.com) Google’s GitHub page says the repository is under active development, and the launch post said additional skills would arrive in the coming weeks and months. (codelabs.developers.google.com) The next public signal is likely to appear first in commits and new folders on the `google/skills` repository, where the current catalog and install instructions are already live. (cloud.google.com 1) (cloud.google.com 2)