agent-browser_skill

Overview

This skill automates browser interactions from the command line to support web testing, form filling, screenshots, video recording, and data extraction. It provides a compact CLI that navigates pages, inspects interactive elements, performs actions by reference or semantic locators, and exports JSON for programmatic parsing. Use it to script reproducible browser workflows and integrate browser automation into AI agents or CI pipelines.

How this skill works

You open a page, run a snapshot to get element refs (e.g., @e1), then perform interactions like click, fill, type, and wait. Snapshots return accessibility trees or filtered interactive elements; repeat snapshots after navigation or major DOM changes to refresh refs. The CLI supports navigation, state/session management, network routing, device emulation, screenshots/PDFs, video recording, and JSON output for automation.

When to use it

• Automated form submission and validation across environments
• End-to-end or regression testing of web applications
• Page scraping and structured data extraction with element refs
• Capturing screenshots, PDFs, or recording interaction videos for demos
• Simulating devices, geolocation, or network conditions during testing

Best practices

• Always run snapshot -i to get stable interactive refs before performing actions
• Resnapshot after navigation or significant DOM updates to avoid stale refs
• Use --session to isolate parallel browser sessions and preserve state
• Combine wait commands (by element, URL, or networkidle) to make flows deterministic
• Use --json for machine-readable output when integrating with other tools

Example use cases

• Fill and submit a login form, save session state, and reuse it in later runs
• Run a capture workflow that extracts text, takes full-page screenshots, and saves a PDF
• Record a demo: explore a site, start record, perform actions, then stop to produce a webm
• Intercept API requests to mock or block third-party resources during tests
• Run accessibility or visual checks by querying computed styles and bounding boxes