Build and run Gemini 2.5 browser-control agents using Playwright to automate complex web-based tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install gemini-computer-use
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install gemini-computer-use using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Gemini Computer Use?
Gemini Computer Use is a sophisticated framework designed to create autonomous agents capable of controlling a web browser just like a human. By leveraging the Gemini 2.5 model alongside Playwright, this skill enables a continuous feedback loop where the agent observes the screen, decides on an action, and executes it in real-time. This integration within the Openclaw Skills ecosystem provides developers with the tools to handle visual-first web navigation and complex UI interactions.
The skill is particularly effective for scenarios where traditional API-based automation is impossible. It uses a structured agent loop that transforms visual screenshots into actionable function calls, ensuring that even the most dynamic web environments can be navigated accurately and safely. With built-in support for different browser channels and safety confirmations, it offers a robust solution for modern web automation needs.
Gemini Computer Use Use Cases
- Automating research tasks that require navigating multiple websites and capturing visual data.
- Performing end-to-end regression testing on web applications without writing manual scripts.
- Interacting with legacy web platforms or SaaS tools that do not provide public APIs.
- Building autonomous shopping or booking assistants that require complex UI navigation.
- Creating visual data scrapers that can bypass challenges inherent in traditional HTML parsing.
How Gemini Computer Use Works
- Visual Capture: The agent captures a high-resolution screenshot of the current browser state to serve as the visual input for the model.
- Goal Processing: The user's prompt and the current screenshot are sent to the Gemini 2.5 model to determine the next logical step.
- Action Generation: The model returns function calls representing specific browser actions such as clicking, typing, or scrolling.
- Safety Verification: If an action is flagged as risky, the system triggers a safety decision loop to request user confirmation before execution.
- Playwright Execution: The validated actions are executed within the browser environment via Playwright.
- Response Iteration: The system sends the result, including the new URL and an updated screenshot, back to the model to continue the loop until the task is complete.
Gemini Computer Use Setup
To begin using this skill, first configure your environment by copying the example file and adding your API keys:
cp env.example env.sh
# Edit env.sh with your API key
source env.sh
Next, set up the Python virtual environment and install the necessary Openclaw Skills dependencies:
python -m venv .venv
source .venv/bin/activate
pip install google-genai playwright
playwright install chromium
Run the agent by providing a prompt and a starting URL:
python scripts/computer_use_agent.py \
--prompt "Search for Openclaw Skills documentation" \
--start-url "https://google.com" \
--turn-limit 10
Gemini Computer Use Data Schema & Taxonomy
The skill manages its operational data through a structured sequence of screenshots and function responses:
| Data Type | Description |
|---|---|
| Screenshot | 1440x900 PNG files used for model perception. |
| Function Call | JSON objects defining browser actions like click or type. |
| Environment Config | Shell-based variables for browser channels (Chrome, Edge, Brave). |
| Session Context | An iterative log of screenshot -> action -> function_response cycles. |
| Safety Flags | Configuration parameters that exclude specific risky UI actions. |
Gemini Computer Use Advanced Features
- Multi-browser support allowing selection between Playwright's bundled Chromium, Google Chrome, Microsoft Edge, or custom executables like Brave.
- Human-in-the-loop safety integration which pauses execution for risky actions requiring explicit user confirmation.
- Action exclusion filtering to prevent the agent from interacting with specific UI elements or executing certain command types.
- Configurable viewport management to ensure visual consistency across different automated sessions.
- Turn-limit controls to manage API consumption and prevent infinite loops in autonomous browsing tasks.
SKILL.md
Loading
Related Openclaw Skills
Context7 Documentation Fetcher
A documentation retrieval skill that connects AI agents to the Context7 API for real-time library specifications and best practices.
Robotic Control
A comprehensive robotic control interface for 6-DOF arms and grippers using voice or code.
HARPA Grid Browser Automation
Automate real web browsers to scrape data, search the web, and execute AI commands directly on live pages.
Moltfounders
A specialized marketplace for AI agents to form teams, recruit collaborators, and manage project workflows through secure API-driven communication.
Get You Some Britches
A comprehensive apparel acquisition framework designed to find the perfect pair of pants when your personal life or wardrobe needs a technical upgrade.
Morning Email Rollup
An automated morning digest that uses AI to summarize your most important emails and list daily calendar events directly to Telegram.
Featured*
