Is a graphical browser window required to use this skill?

No, it runs in headless mode by default. However, you can use the --headed flag to launch a visible browser window for debugging purposes.

How does the ref-based selection work?

By running a 'snapshot' command, the tool generates a list of interactive elements assigned to unique IDs like @e1 or @e2. You can then perform actions like click or fill directly on those refs.

Can I run multiple browser sessions at once?

Yes, Agent Browser supports independent parallel sessions using the --session flag, allowing you to manage different websites or workflows simultaneously.

What makes Agent Browser different from Playwright?

While Playwright is a comprehensive framework, Agent Browser is a CLI-first tool optimized for AI agents using a ref-based system (@e1, @e2) that simplifies element selection within terminal environments.

What formats can I export page data into?

You can export data as JSON via snapshots, capture visual screenshots in PNG format (viewport or full-page), or save the entire page as a PDF.

Agent Browser Automation

Name: Agent Browser Automation
Author: EveryInc

byEveryInc

•

5,347

ブラウザ自動化

Automates web browser interactions and data extraction using a command-line interface optimized for AI agents.

概要

Agent Browser is a specialized skill for Claude Code that enables seamless web automation through Vercel's agent-browser CLI. It simplifies complex web tasks by using a unique ref-based selection system from accessibility snapshots, allowing Claude to navigate pages, fill forms, and scrape data without traditional selector fragility. This skill is ideal for developers who prefer bash-based workflows for quick automation, site testing, and capturing visual data through screenshots or PDFs directly within their terminal environment.

主な機能

Semantic locators for finding elements by role, text, or label
Ref-based element interaction using accessibility snapshots (@e1, @e2)
5,347 GitHub stars
High-fidelity data extraction with JSON-formatted snapshots
Full headless browser control including navigation and session management
Visual output support for viewport and full-page screenshots or PDFs

ユースケース

Generating full-page screenshots and PDFs for documentation or debugging
Automated form filling and login flow verification
Web scraping and content extraction from dynamic JavaScript-heavy sites

概要

主な機能

Semantic locators for finding elements by role, text, or label
Ref-based element interaction using accessibility snapshots (@e1, @e2)
5,347 GitHub stars
High-fidelity data extraction with JSON-formatted snapshots
Full headless browser control including navigation and session management
Visual output support for viewport and full-page screenshots or PDFs

ユースケース

Generating full-page screenshots and PDFs for documentation or debugging
Automated form filling and login flow verification
Web scraping and content extraction from dynamic JavaScript-heavy sites