agent-browser_skill

Overview

This skill provides a CLI-driven browser automation toolkit for web testing, screenshots, data extraction, and demo recording. It exposes navigation, page inspection, DOM interactions, and recording features to automate workflows and reproduce user journeys. Use it to script tests, capture visual evidence, and extract structured page data quickly.

How this skill works

It opens a real browser context and exposes commands to navigate, snapshot the accessibility/DOM tree, and interact with elements using stable refs or semantic locators. Snapshots return element refs like @e1 which you reuse for clicks, fills, checks, and queries. Recording, network routing, and session state let you reproduce flows or run parallel isolated sessions.

When to use it

• Automated end-to-end testing of web apps and form flows
• Capture screenshots, full-page PDFs, or video demos of pages
• Extract text, attributes, or counts from pages for scraping
• Reproduce and debug UI bugs with headed mode and console logs
• Simulate user interactions: clicks, typing, file uploads, drag-and-drop

Best practices

• Run snapshot -i to gather interactive element refs before interacting
• Re-snapshot after navigation or major DOM updates to refresh refs
• Use semantic locators (role/name/text) for more resilient targeting
• Save/load session state for authenticated flows instead of re-login
• Pipe --json output into scripts for reliable machine parsing

Example use cases

• Automate a sign-in test: open page, fill email/password, click submit, wait for dashboard
• Record a product walkthrough: start record, navigate, click features, stop to save webm
• Take full-page screenshots for visual regression or design review
• Scrape a listing page: snapshot, get text/@href for each item, export counts
• Mock API responses during tests with network route and --body to simulate edge cases