agent-browser-0.2.0_skill

Overview

This skill provides a fast, Rust-based headless browser automation CLI with a Node.js fallback that lets AI agents and scripts navigate, inspect, interact with, and capture web pages using structured commands. It exposes snapshot refs and semantic finders so agents can reliably click, type, upload, screenshot, record, and query page state programmatically. The CLI is optimized for automation, reproducible sessions, and machine-readable JSON output.

How this skill works

The tool opens a browser context, captures an accessibility-style snapshot that returns element references (e.g., @e1), and uses those refs for direct interactions like click, fill, hover, and upload. It supports semantic locators (role, text, label) as alternatives to refs, and provides commands for navigation, waiting, network routing, storage, and recording. Use --json for machine-readable responses and sessions or saved state to persist cookies and storage between runs.

When to use it

• Automating end-to-end flows (logins, form submissions, checkout) for agents
• Scraping or extracting structured content where element refs simplify logic
• Creating reproducible demos via screenshot, PDF, or video recording
• Testing interactions across isolated sessions in parallel
• Intercepting or mocking network requests during automated tasks
• Debugging client-side behavior with headed mode, console, and traces

Best practices

• Always run snapshot -i after navigation or significant DOM changes to obtain fresh refs
• Prefer fill over type when replacing input contents to avoid leftover text
• Use --json for downstream parsing and stable automation logic
• Save and load state for authenticated workflows instead of re-authenticating repeatedly
• Limit snapshot depth or scope with -d and -s to reduce noise and speed up discovery
• Use semantic locators (role/text/label) when refs are brittle across small UI changes

Example use cases

• Log in once, state save, then run periodic data extraction on authenticated pages
• Drive a multi-step checkout flow: open, snapshot -i, fill inputs, click buttons, wait for networkidle
• Record a guided demo: record start, perform interactions, record stop to produce a webm
• Run parallel sessions to test multi-tenant behavior or concurrent user flows
• Intercept API calls to mock responses during automated UI tests