gemini-automation

name	gemini-automation
description	Gemini PRO 3.1 UI Automation via existing Chrome profile. Provides AI agent-accessible methods for prompt sending, deep search, image generation, and video generation using dev-browser extension. Use when: automating Gemini UI, generating images/videos with Gemini, conducting deep research via Gemini, or needing AI agent access to Gemini PRO features.
license	MIT
metadata	{"author":"user","version":"1.4.0"}

Gemini Automation

You are an expert in automating Gemini PRO UI through Chrome's existing browser session using the dev-browser extension and relay server.

Quick Start - Read This First

OPTIONAL: Optimize Your Prompt First

Before running CLI, you can OPTIONALLY use the prompt-optimization skill to improve results:

1. Load prompt-optimization skill
2. Provide your raw prompt
3. Use the optimized prompt in CLI

This uses Google official templates for better image/video/deepsearch results.

---

When user asks to generate image/video/prompt with Gemini:

Step 1: Ensure relay server is running

cd /tmp/dev-browser/skills/dev-browser && npx tsx scripts/start-relay.ts &

Wait 3 seconds, verify with: curl -s http://localhost:9222

Step 2: Run the CLI command

For IMAGE generation:

cd /tmp/dev-browser/skills/dev-browser && npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts image "USER_PROMPT"

For VIDEO generation:

cd /tmp/dev-browser/skills/dev-browser && npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts video "USER_PROMPT"

For PROMPT/CHAT:

cd /tmp/dev-browser/skills/dev-browser && npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts prompt "USER_PROMPT"

For DEEP RESEARCH:

cd /tmp/dev-browser/skills/dev-browser && npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts deepsearch "USER_QUERY"

Step 3: Report results

• Tell user where the screenshot was saved
• Show the output path

---

Prerequisites

1. Chrome Extension Setup

• Install dev-browser Chrome extension
• Set extension mode to "Active" (not "Off")
• Ensure your Google account is logged into Gemini

2. Relay Server (CRITICAL)

The relay server must be running for ANY automation:

# Start relay server (one-time setup)
cd /tmp/dev-browser/skills/dev-browser
npx tsx scripts/start-relay.ts &

Verify it's running:

curl -s http://localhost:9222
# Should return: {"wsEndpoint":"ws://127.0.0.1:9222/cdp","extensionConnected":true,"mode":"extension"}

---

Available Actions

1. Send Prompt

Command: npx tsx scripts/cli.ts prompt "your message"

What it does:

• Connects to Chrome via relay server
• Switches to PRO model if needed
• Types prompt into textbox
• Presses Enter to send
• Waits for response (15s)
• Saves screenshot to {outputDir}/gemini-response.png

2. Deep Search

Command: npx tsx scripts/cli.ts deepsearch "your research query"

What it does:

• Opens Gemini Deep Research mode
• Enters query and submits
• Auto-clicks approve button when research plan appears
• Waits for research to complete (2-3 minutes)
• Saves full research text to {outputDir}/gemini-deep-research.txt
• Saves screenshot to {outputDir}/gemini-deepsearch.png

IMPORTANT: Deep Research takes 2-3 minutes. Be patient!

Example:

cd /tmp/dev-browser/skills/dev-browser
npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts deepsearch "AI trends 2026"

After completion, read the research from {outputDir}/gemini-deep-research.txt

3. Image Generation

Command: npx tsx scripts/cli.ts image "sunset over ocean"

What it does:

• Sends prompt to Gemini
• Waits for image creation (25s)
• Saves full-page screenshot to {outputDir}/gemini-image.png

Example:

cd /tmp/dev-browser/skills/dev-browser
npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts image "Okyanusun üzerinde güneş batışı"

4. Video Generation

Command: npx tsx scripts/cli.ts video "space travel"

What it does:

• Sends prompt for video
• Waits for video creation (30s)
• Saves screenshot to {outputDir}/gemini-video.png

---

Configuration

Output Directory

By default, files are saved to user's home directory. You can change this:

# Set custom output directory
GEMINI_OUTPUT_DIR=/tmp/output npx tsx scripts/cli.ts image "cat"

# Or export for multiple commands
export GEMINI_OUTPUT_DIR=/tmp
npx tsx scripts/cli.ts image "cat"
npx tsx scripts/cli.ts prompt "hello"

Timeout Configuration

You can customize timeouts via environment variables:

# Page load timeout (default: 30000ms)
GEMINI_TIMEOUT_PAGE_LOAD=60000 npx tsx scripts/cli.ts prompt "hello"

# Response wait timeout (default: 20000ms)
GEMINI_TIMEOUT_RESPONSE=30000 npx tsx scripts/cli.ts prompt "complex task"

# Deep search timeout (default: 180000ms = 3 minutes)
GEMINI_TIMEOUT_DEEP_SEARCH=300000 npx tsx scripts/cli.ts deepsearch "AI"

# Image generation timeout (default: 25000ms)
GEMINI_TIMEOUT_IMAGE=35000 npx tsx scripts/cli.ts image "detailed scene"

# Video generation timeout (default: 30000ms)
GEMINI_TIMEOUT_VIDEO=60000 npx tsx scripts/cli.ts video "animation"

---

Error Handling

Error	Solution
"Connection refused"	Start relay server: cd /tmp/dev-browser/skills/dev-browser && npx tsx scripts/start-relay.ts &
"Extension not connected"	Set dev-browser extension to "Active" mode
"PRO model not available"	Check if user has PRO access
"Element not found"	Gemini UI may have changed - script will try fallback selectors
"Output directory does not exist"	Set GEMINI_OUTPUT_DIR to a valid path
"Prompt cannot be empty"	Provide a non-empty prompt
"Prompt exceeds maximum length"	Prompt is too long (max 10000 chars)

Troubleshooting

Relay server won't start:

1. Check if port 9222 is available: lsof -i :9222
2. Kill existing process if needed
3. Restart: cd /tmp/dev-browser/skills/dev-browser && npx tsx scripts/start-relay.ts &

Extension shows as not connected:

1. Open Chrome and navigate to any page
2. Click the dev-browser extension icon
3. Ensure mode is set to "Active"

Operation times out:

• Increase timeout values using environment variables
• Deep research may take longer depending on query complexity

---

Output Format

When completing a Gemini automation task, report:

## Action Taken
[What was done]

## Connection Status
✅ Connected to existing Chrome session / ❌ Connection failed

## Model
PRO model active / Still on Hızlı

## Result
[Response summary / Error details]

## Files Generated
[screenshot paths]

---

Dependencies

• dev-browser Chrome extension (installed and active)
• Relay server running on localhost:9222
• Existing Gemini login session
• PRO model access (user account dependent)

---

Example Workflow

1. User says: "Generate an image of a cat"
2. (OPTIONAL) Use prompt-optimization skill to improve the prompt
3. Check if relay server is running
4. If not, start it: cd /tmp/dev-browser/skills/dev-browser && npx tsx scripts/start-relay.ts &
5. Run: cd /tmp/dev-browser/skills/dev-browser && npx tsx /Users/user/.claude/skills/gemini-automation/scripts/cli.ts image "cat"
6. Report: "✅ Image generated: ~/gemini-image.png"

---

API Reference

Module Import

import { geminiAI } from './cli.js';

// Generate image
await geminiAI.image("sunset over ocean");

// Generate video  
await geminiAI.video("space travel");

// Send prompt
await geminiAI.prompt("hello world");

// Run deep search
await geminiAI.deepSearch("AI trends 2026");

// Disconnect when done
await geminiAI.disconnect();

Class Methods

const gemini = new GeminiAutomation();

await gemini.connect();           // Connect to Chrome
await gemini.prompt("text");      // Send prompt
await gemini.image("prompt");      // Generate image
await gemini.video("prompt");      // Generate video
await gemini.deepSearch("query"); // Deep research
await gemini.disconnect();        // Cleanup

---

Version History

• 1.4.0: Added prompt-optimization skill reference
• 1.3.0: Added input validation, configurable timeouts, proper TypeScript types, input sanitization, fixed race condition
• 1.2.0: Added configurable output directory, improved error handling, localization fallback selectors
• 1.1.0: Initial release with basic prompt, image, video, and deep search support