browser-automation

What This Skill Does

The browser-automation skill provides a robust interface for controlling Chrome programmatically via PinchTab. Designed for AI agents, it bridges the gap between high-level task objectives and low-level browser execution. By interacting directly with the accessibility tree rather than relying on bloated visual screenshots, the skill achieves massive token efficiency, allowing agents to navigate complex web apps while keeping context windows clean. Whether you need to scrape dynamic content, validate form submissions, or automate multi-step workflows, this skill handles the heavy lifting of process orchestration, headless/headed state management, and tab lifecycle control.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/huamu668/browser-automation-pin

Ensure you have the PinchTab server installed on your system or running via Docker, as the agent relies on the local HTTP API at localhost:9867 to bridge commands into the browser environment.

Use Cases

• Web Scraping: Extracting data from dynamic, JavaScript-heavy sites that standard crawlers cannot interpret.
• Quality Assurance: Automated end-to-end testing for web applications by simulating user clicks, text input, and navigation.
• Agentic Research: Browsing the internet to gather news, monitor site changes, or verify information as part of a larger research task.
• Form Automation: Programmatically filling out registrations, surveys, or internal business tools that lack APIs.

Example Prompts

1. "Open a headless browser, navigate to example.com, wait for the page to load, and extract all the links found in the main content area."
2. "Log into the portal using my work profile, fill out the user email field, click the submit button, and let me know if the confirmation text appears."
3. "Start a new headed instance, go to the tech news site, and summarize the top three headlines from the homepage."

Tips & Limitations

• The 3-Second Rule: Always allow a 3-second buffer after navigation. Chrome needs this time to render the accessibility tree; attempting to snap or extract elements immediately will result in an empty response.
• Token Efficiency: Use the suggested JQ filtering patterns to strip out non-essential elements like navigation bars or footers before passing the data to the LLM.
• Persistence: Use browser profiles for tasks requiring authentication so you do not have to perform login sequences every time the instance restarts.
• Headless vs. Headed: Use headless mode for background tasks to save system resources; use headed mode only when you need to visually debug or interact with sites requiring MFA or complex CAPTCHAs.