PlugWiki

gemini-interactions-api Skill (Google) for Claude: What It Does and How to Install It

Guides Claude to use Google's stateful Gemini Interactions API for multi-turn chats, streaming, structured output, tools and agents.

Installs
10.9K
on skills.sh
Stars
21.8K
on GitHub
Files
3
besides SKILL.md
Updated
Nov 2026
last repo commit

This skill teaches Claude to sign in to, connect to and call the Gemini Interactions API on Gemini Enterprise Agent Platform. It covers client setup in Python and TypeScript, single and multi-turn calls, streaming, structured output, function calling, background agents and plain REST calls.

It needs a Google Cloud project, a gcloud sign-in and the Google Gen AI SDK (google-genai 2.3.0 or later for Python, @google/genai 2.3.0 or later for JavaScript). SKILL.md also tells Claude to tag gcloud commands and API calls with a usage-attribution string, described in the safety section.

Good for

  • Developers building on Google's Gemini Enterprise Agent Platform
  • Teams adding multi-turn or background agent runs to an app
  • Developers using structured output and function calling
  • Anyone calling the API from curl or shell scripts

Install gemini-interactions-api

Claude Code

  1. 1

    Copy this prompt.

    Install the gemini-interactions-api skill in Claude Code. Run this command:

npx skills add https://github.com/google/skills --skill gemini-interactions-api

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

Pick one route. The skills CLI installs just this skill; the repo README also shows npx skills add google/skills for choosing from all of its skills. The skill needs a Google Cloud project, the gcloud tool and the Google Gen AI SDK for Python or JavaScript, so the box has no Claude app upload.

Built from the package data we hold. Not yet tested by us.

What's in the skill

When it loads

Guides the usage of Gemini Interactions API on Gemini Enterprise Agent Platform. Use when the user wants to use the stateful, server-managed Interactions API for multi-turn conversations, background execution, streaming, structured output, and function calling on the Agent Platform.

Sections in SKILL.md

  1. Attribution
  2. 1. Authentication
  3. 2. Client Initialization
  4. Option A: Environment Variables (Recommended)
  5. Option B: Explicit Inline Parameters
  6. Option C: Express Mode (API Key)
  7. 3. Core Interactions API Usage
  8. Quick Start (Single-Turn)
  9. Stateful Conversation (Multi-Turn)
  10. Real-Time Streaming
  11. Structured Output (Pydantic / Polymorphic response_format)
  12. Function Calling (Agent Tool Use)
  13. Agents & Long-Running Tasks
  14. 4. Accessing the Interactions API via REST
  15. 5. Data Model & Step Types Reference

Key facts

Author
google (repo owner)
License
Apache-2.0 (the README says Apache 2.0)
Installs on skills.sh
10,894 (Nov 2, 2026)
GitHub stars
21,764 for the whole google/skills repo (Nov 2, 2026)
Last repo commit
Nov 1, 2026
What it includes
4 files including SKILL.md: references/attribution.md, data_model.md and rest_api.md (no scripts)
Needs
A Google Cloud project, gcloud (Application Default Credentials) or an Express Mode API key, and the Google Gen AI SDK
Skill version
2.0.1 (SKILL.md metadata)

What it includes

  • Sign in with gcloud and enable the aiplatform API

    Runs gcloud auth application-default login, then gcloud services enable aiplatform.googleapis.com if the API is not on yet.

    Runs code · Authentication

  • Set up the SDK from env vars, inline settings or an API key

    Three options: environment variables, explicit project and location, or Express Mode with an API key.

    Write · Client initialization

  • Create an interaction and continue it by ID

    Interactions are stored by default and continued with previous_interaction_id.

    Try: “Write a Python script that asks Gemini a question, then follows up in the same interaction.”

    Write · Single-turn and multi-turn

  • Read typed events as the model writes

    Shows the event order from interaction.created to interaction.completed.

    Write · Streaming

  • Return JSON that matches a Pydantic or JSON schema

    Passes a schema as response_format so the output text is valid JSON.

    Write · Structured output

  • Let the model request your local functions

    Passes tools, reads function_call steps, and sends back a function_result step.

    Write · Function calling

  • Managed agents, Deep Research and custom agents

    Targets antigravity-preview-05-2026, deep-research-preview-04-2026 or custom agents, with background=True and polling.

    Try: “Run a deep research agent on solar energy providers in the background.”

    Network · Agents and background runs

  • Call the endpoint with curl

    Documents the interactions endpoint on aiplatform.googleapis.com, with curl examples in references/rest_api.md.

    Runs code · REST access

  • Step types, content types, events and 7-day storage

    references/data_model.md lists step and event types and the retention rules.

    Read · Data model reference

  • Tag gcloud and HTTP calls with the skill name

    references/attribution.md says to prefix terminal gcloud commands and add a User-Agent suffix; generated code is not tagged.

    Runs code · Usage attribution

From the official docs

Permissions

Medium risk

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • It can enable an API in your Google Cloud project

    Step 1 runs gcloud services enable aiplatform.googleapis.com if the API is not already on. Check the active project before you let Claude run it. github.com

  • Interactions are stored for 7 days by default

    data_model.md says interactions are stored with store=True and retained for 7 days. Passing store=False turns retention off, and also turns off previous_interaction_id and background runs. github.com

  • Express Mode puts an API key in client code

    Option C in SKILL.md passes api_key to the client. Read the key from an environment variable instead of writing it into source. github.com

  • Managed agents run in a remote sandbox

    The antigravity-preview-05-2026 agent is described as having code execution, file management and web browsing in a sandboxed Linux environment. Only use it for work you are happy to run there. github.com

  • Terminal commands carry a usage tag

    attribution.md says to prefix gcloud commands with CLOUDSDK_METRICS_REQUEST_ATTRIBUTION and add a User-Agent suffix. It says the tag carries no user data and that the tags are omitted if you ask. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details
Turning on read-onlyNo read-only mode. Use version control and a test project.
Talks toaiplatform.googleapis.com
Sign-in and keysUses Application Default Credentials from gcloud auth application-default login, or an API key in Express Mode. SKILL.md does not say where either is kept.
Scopes

  • Claude can run gcloud auth application-default login and gcloud services enable aiplatform.googleapis.com

  • Claude can send prompts and tool results to the Interactions API in your Google Cloud project

  • Managed agents can run code, manage files and browse the web in a sandbox

Questions

What does the gemini-interactions-api skill do?

It guides Claude through Google's Gemini Interactions API on Gemini Enterprise Agent Platform. Claude uses it to set up auth and a client, then write multi-turn, streaming, structured-output, tool-calling and agent code.

Is this skill from Google?

Yes. It is in google/skills, Google's own repository, which the README describes as Agent Skills for Google products and technologies. SKILL.md carries its own version, 2.0.1.

How do I install gemini-interactions-api?

Use the install box at the top of this page. It has no Claude app upload because the skill needs gcloud and an SDK on your machine. Claude Code and Codex are covered.

Do I need a Google Cloud account or API key?

You need one of two. SKILL.md has you sign in with Application Default Credentials and enable the aiplatform API in a Google Cloud project, or use Express Mode with an API key.

Which models does it use?

SKILL.md names gemini-3.8-flash as the recommended default, gemini-3.1-pro-preview for complex reasoning and gemini-3.5-flash-lite for cheap, frequent tasks. It tells Claude to warn that gemini-2.5, 2.0 and 1.5 models are deprecated and unsupported for Interactions.

Does it send usage data to Google?

It adds a tag. references/attribution.md says to prefix terminal gcloud commands with CLOUDSDK_METRICS_REQUEST_ATTRIBUTION and add a User-Agent suffix, and that the tag "identifies only the skill and its version; it carries no user data." If you ask to remove the tags, it omits them.

Is it safe?

It bundles no scripts. When you ask for code, Claude may run gcloud login and service-enable commands against your Google Cloud project, so check which project is active first. Interactions are stored for 7 days unless you pass store=False.

Information

updated Nov 2, 2026

Repository
google/skills
Folder
skills/cloud/gemini-interactions-api

google/skillsskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5
  1. gemini-interactions-api SKILL.md · github.com
  2. google/skills README · github.com
  3. attribution.md · github.com
  4. data_model.md · github.com
  5. rest_api.md · github.com

Updated Nov 2, 2026.