# Gemini

> Google Gemini API Skill

- Skill: `ashish7802/gemini` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx skillmds@latest add ashish7802/gemini`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ashish7802/gemini/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Integrations & APIs
- Author: ashish7802 (https://skillmd.com/u/ashish7802)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/ashish7802/gemini

---

# Google Gemini API Skill

## Overview
Google Gemini offers natively multimodal capabilities (text, image, audio, video). This skill focuses on `@google/genai` and the v1beta API surface.

## Installation
```bash
npm install @google/genai
pip install google-genai
```

## Authentication
Use a Google API key passed to the client initialization. For GCP environments, Vertex AI authentication via IAM is preferred.

## Core Concepts
- **Parts**: The building blocks of a message. A part can be text or inline data (images).
- **Gemini 1.5 Pro**: Features a massive 1-million to 2-million token context window.
- **System Instructions**: Provided at the model initialization level.

## Common Workflows
1. Initialize `GoogleGenAI`.
2. Call `models.generateContent` with a multimodal array of parts.
3. Extract the text from the response candidate.

## Error Handling
Watch for `FinishReason.SAFETY`. If the model refuses to answer due to safety settings, the response will be empty but the `finishReason` will indicate why.

## Security
Tune safety settings (`HARM_CATEGORY_HATE_SPEECH`, etc.) according to your application's risk tolerance.

## Rate Limits
Free tier offers 15 RPM for Gemini 1.5 Flash. Paid tiers depend on GCP quotas.

## Best Practices
Leverage the massive context window by uploading entire codebases or PDFs rather than aggressively chunking, as Gemini 1.5's recall is exceptionally high.

## Troubleshooting
If multimodal requests fail, ensure inline data is base64 encoded and the correct MIME type (e.g., `image/jpeg`) is specified.

## References
- [API Reference](https://ai.google.dev/api)

## Why use this skill
Use this when your agent works with **gemini** — structured patterns beat pasted docs and prevent common hallucinations.

## AI pitfalls
- Using deprecated model IDs or wrong API endpoints
- Confusing chat vs completions vs embeddings APIs
- Omitting rate-limit and token budget handling

## Production checklist
- [ ] Secrets in environment variables, not source code
- [ ] Error handling and logging in place
- [ ] Rate limits and timeouts configured

## Related skills
- [`langchain`](../langchain/SKILL.md) — integrates with
- [`llamaindex`](../llamaindex/SKILL.md) — integrates with

---
> **Last Verified:** 2026-07-02