Start ADK from official docs in your preferred language

Who it is for

This skill fits people who want one open-source, code-first toolkit to build, evaluate, and deploy complex AI agents. The official name is Agent Development Kit（ADK）. It also fits people who already have a preferred programming language and are ready to start from the official “preferred language” entry. The source says you can use ADK in your preferred language, but this passage does not give concrete language addresses.

It does not fit treating this repository as an install command. This is a documentation repository. ADK is optimized for Gemini and the Google ecosystem, but is written as model-agnostic and deploy-agnostic, and compatible with other frameworks. If you only want to read docs first and do not yet want containerized deploy, do not start with Cloud Run or GKE.

The official repository is https://github.com/google/adk-docs Operate with notes already written on that page. Do not invent an installer or a documentation address that never appeared.

First attempt in 3 minutes

Follow only this official-docs path: first pick your preferred language, then find the matching start in the official notes, instead of running a package-install command that never appeared.

Prerequisite: you can open the official documentation repository and already have a language you intend to use. The first time, do not deploy, and do not write secrets into the conversation.

Install location: open https://github.com/google/adk-docs Follow “use ADK in your preferred language” to find the official start. The source does not write pip or npm commands; do not invent them.

Low-risk task: have the agent only read and summarize three facts — it is a code-first toolkit; though optimized for Gemini and the Google ecosystem, it is model-agnostic and deploy-agnostic; and it provides optimized documentation context compatible with AI code editors. Do not let it invent documentation file addresses, and do not let it deploy to Cloud Run, GKE, or Agent Runtime.

Success signal: you can point out that the next step must return to official docs and start in your own language, and no invented install command was run. If the agent gives a URL or package name that never appeared, stop.

Usage cases

Case 1: Start making an agent in a language you already know. The source says you can use a preferred language. It fits people who first pick a language, then only open the official start page. If the concrete language path does not appear in the current notes, stop at comparing the repository directory; do not invent a get-started address.

Case 2: Give an AI code editor optimized documentation context. The source says ADK provides optimized documentation context compatible with AI code editors, to speed development. It fits people already using an editor who only want to confirm they should load official optimized docs, not paste a random blog. When filenames and addresses are not listed in the source, do not invent them.

Boundaries

This repository is documentation, not an installer. Do not write a package-manager command that never appeared into the main path. Capability coverage is build, evaluate, and deploy. On deploy it only writes that you can containerize and put it on Cloud Run or GKE, or extend with Agent Runtime. Those all have side effects and cannot be the first attempt.

It is optimized for Gemini and the Google ecosystem, but officially it is also model-agnostic, deploy-agnostic, and compatible with other frameworks. Do not write it as “Gemini only.” Optimized documentation context is compatible with AI code editors, but there is no client list. The official page is https://github.com/google/adk-docs If an address or command an agent suggests disagrees with that page, treat that page as authoritative.