Verified:GitHub

gemini-computer-use

Automates browser tasks using Gemini Computer Use agents and Playwright.

1.0k|56|Updated Jan 16, 2026

Search Tags:#AI automation #browser automation #Playwright #web navigation #Gemini Computer Use #agent loop #safety confirmation

Ready to run

npx skills add https://github.com/am-will/codex-skills --skill gemini-computer-use

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: gemini-computer-use
Source: https://github.com/am-will/codex-skills/tree/main/skills/gemini-computer-use
Command: npx skills add https://github.com/am-will/codex-skills --skill gemini-computer-use

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires playwright, google-generativeai, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill enables building and running Gemini 2.5 Computer Use browser-control agents with Playwright to automate repetitive web browsing tasks. It supports an agent loop where the model suggests a function_call and the agent executes actions, then returns a function_response to continue the workflow, including optional safety confirmations for risky UI actions.

Core Features & Use Cases

Gemini Computer Use agents controlled by Playwright for browser automation.
Deterministic action execution with a loop (screenshot → function_call → action → function_response).
Safety confirmation prompts for high-risk UI actions to prevent unwanted changes.
Real-time feedback via screenshots and URL context after each action.

Quick Start

Tell the agent to visit example.com, take a screenshot, and return the page title.

Frequently Asked Questions about gemini-computer-use

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automate browser tasks using Gemini Computer Use?▼

You can automate browser tasks using Gemini Computer Use by running a Python agent loop with Playwright that takes a screenshot, processes a function_call, executes the action, and returns a function_response. This loop handles web navigation and UI testing deterministically.

How does the agent loop work for web navigation with Playwright?▼

The agent loop works by capturing a screenshot and URL context, sending them to the Gemini model to suggest a function_call, executing that browser action via Playwright, and returning a function_response to continue the workflow. This provides real-time feedback after each step.

Do I need a Python runtime and Playwright to run Gemini Computer Use agents?▼

Yes, you need a Python runtime, the Playwright library, the google-generativeai dependency, and environment configuration for browser control. Access to the Gemini Computer Use model is also required to execute the automation workflows.

Can I add safety confirmations for risky UI actions in browser automation?▼

Yes, you can enable optional safety confirmation prompts before executing high-risk UI actions in browser automation. This prevents unwanted changes by requiring manual approval during the agent loop when risky operations are detected.

What is the best way to verify content across pages with AI automation?▼

The best way to verify content across pages with AI automation is using a deterministic agent loop that screenshots pages, executes actions, and returns responses. This approach applies to data gathering and verifying content across pages safely.

Why use Playwright for AI automation instead of other browser automation tools?▼

Playwright provides deterministic browser control that integrates directly with the Gemini Computer Use agent loop. It executes function_calls for web navigation and returns function_responses, offering real-time screenshot feedback for reliable automated workflows.