Install
npx skills add google-gemini/gemini-skills --skill gemini-api-devWho it's for
Use this skill library when you want an agent to write call code for the Gemini API, SDKs, and model interactions. The source describes it as a skill library for Gemini API, SDK, and model interactions, and specifically names gemini-api-dev: application development against the Interactions API, covering text generation, multi-turn conversation, streaming output, function calling, structured output, image generation, Deep Research agents, guardrails for deprecated models, and the Python and TypeScript SDKs. Docs:
https://ai.google.dev/gemini-api/docs/interactions?ua=chat
Official repository:
https://github.com/google-gemini/gemini-skills
The source says you can install these skills into the AI coding assistants and package managers you already use, but it does not name specific products. Do not invent a client list, and do not write it as exclusive to a few environments. If you are not ready to run the official npx command on this machine, do not continue yet.
The source also states: this is not an officially supported Google product, and it cannot join the Google Open Source Software Vulnerability Reward Program. Evaluation numbers only show that after the skill is installed, the rate at which the agent generates correct API code changes. Do not treat that as the success criterion for one call on your machine. Do not infer that the skill can only be used on the two models mentioned in the evaluation. The library may contain other skills, but the source only writes gemini-api-dev as a copyable install example. The first time, do not name skills that never appeared.
First try in 3 minutes
First open the official repository and check:
https://github.com/google-gemini/gemini-skills
The recommended path is the official npx skills add. The first time, browse only; do not install a pile of skills yet. Browsing is interactive; before you run it, confirm you are only viewing the list:
npx skills add google-gemini/gemini-skills --list
After you see the list, the concrete example the source gives is gemini-api-dev. Install writes skill files on this machine; before you run it, confirm you want only this one:
npx skills add google-gemini/gemini-skills --skill gemini-api-dev
Do not change it to a skill name that never appeared, and do not invent a second installer. The source does not write a skill landing directory; do not invent a path.
The first time in the agent, only check; do not call a live API yet, and do not generate images or start Deep Research. Ask the agent to state which Interactions API capabilities gemini-api-dev covers, and that this is not an officially supported Google product. The observation is that it can recite text generation, multi-turn conversation, streaming output, function calling, structured output, and the Python and TypeScript SDKs, and has not gone online yet.
Writing real call code goes online and may consume quota. Continue only when you are ready to use your own Gemini credentials; the source does not give a key field name, so do not write a placeholder key. The first time, do not touch image generation or Deep Research.
Use cases
Case 1: Browse first, then install one skill. Run npx skills add google-gemini/gemini-skills --list, confirm you can see skills in the list, then install gemini-api-dev. Fits people who only want to attach the Interactions API skill to an agent and are not ready to send requests. Success signal: list can open, and the install command for the named skill ends normally. The source does not write the full fields of list; do not invent other --skill names.
Case 2: First check the wiring scope, then decide whether to write code. After installing gemini-api-dev, have the agent recite that it covers text generation, multi-turn conversation, streaming output, function calling, structured output, image generation, Deep Research agents, deprecated-model guardrails, and the Python and TypeScript SDKs. Fits people who worry the agent will go online or generate images immediately. Success signal: it can state the scope clearly and admit this is not an officially supported product. When you need a real call, first confirm this run will access the Gemini API and prepare credentials yourself; the source gives no sample request parameters, so do not invent them. Do not treat the Gemini 3 Flash and Gemini 3.1 Pro numbers in the evaluation as a benchmark you must reproduce.
Boundaries
The main path is npx skills add google-gemini/gemini-skills, browse with --list, and the example skill is --skill gemini-api-dev. Do not invent other skill names or installers. The source says you can install into common AI coding assistants and package managers, but it does not name products; do not add a roster, and do not write it as an exclusive scope.
This is not an officially supported Google product. Vulnerability reward program notes:
https://bughunters.google.com/open-source-security
Capability stops at the Interactions API items above and the two SDKs. Image generation and Deep Research have external side effects; do not do them the first time. The source does not write key fields, quota, or uninstall steps. Do not fill in unwritten model names, client names, or request parameters. First-time success is only the browse command or install finishing, plus the agent reciting the scope; before going online you must confirm credentials and that this run will access the API. The source does not write uninstall or update commands; do not invent a second installer. If the agent starts generating images or launching Deep Research, stop first and return to read-only explanation. Evaluation percentages cannot replace your own one-time success signal.