google-gemini-api

What it does

Instructs the agent to integrate the Gemini API using the current @google/genai SDK (not the deprecated @google/generative-ai), enabling text generation, multimodal inputs (images, video, audio, PDFs), function calling, thinking mode, and streaming. It also notes a 1M input token context and guidance to prevent 14 documented errors, with use cases centered on Gemini integration, multimodal AI, and thinking mode for reasoning enhancements.

How it works

The guide provides a complete end-to-end usage pattern including:

• Installation instructions for the correct SDK version (npm install @google/genai@1.34.0) and a fallback warning against the deprecated SDK.
• Environment setup for GEMINI_API_KEY.
• Code samples for both Node.js SDK and fetch-based approaches for:
  • Basic Text Generation (generateContent) and streaming (generateContentStream) workflows.
  • Streaming via SSE with explicit guidance on endpoint, request structure, and client-side parsing.
  • Multimodal inputs across images, video, audio, and PDFs using inlineData with base64-encoded media and MIME types.
  • Function calling (tool integration) with function declarations and call flows.
• Details on current Gemini models, including 2.5 series (1,048,576 token context, 65,536 output tokens) and 3.x series capabilities like thinking mode, streaming, and function calling.
• Notes on avoiding common mistakes (e.g., token context windows) and distinctions between SDK and fetch approaches.
• Example payloads for quick-starts, streaming, and multimodal content processing.

When to use it

Use when you need Gemini integration, multimodal AI capabilities, and thinking mode for reasoning tasks. Use when troubleshooting SDK deprecation, model discovery, context window concerns, function calling errors, streaming integrity, safety settings, or rate limits.

What it can touch

The skill declares and demonstrates usage with:

• Node.js SDK: @google/genai (Example imports: import { GoogleGenAI } from '@google/genai')
• Fetch-based REST calls (endpoints like https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent and :streamGenerateContent)
• Media content handling via inlineData with mimeType and base64-encoded payloads (images, video, audio, PDFs)
• Function calling declarations and usage (e.g., get_current_weather as a tool)

Caveats

• License: MIT (as declared in the skill metadata)
• It emphasizes using the non-deprecated SDK (@google/genai), with a migration note from @google/generative-ai and a warning about deprecated code in practice.
• The guide contains model-specific context limits and feature availability per model (e.g., Gemini 2.5 vs Gemini 3.x), and it warns about preview models and SLAs.
• It explicitly documents known issues and error-handling sections, including rate limits and safety configurations.