Agent Browser — Core

What it does

Provides the core command reference and patterns for agent-browser, a compact Chrome/Chromium automation CLI tailored for AI agents. It teaches the snapshot-and-ref workflow, how to read pages, interact with elements (click, fill, type, select), manage tabs and sessions, capture screenshots and videos, and handle waits, auth, and common failures. Concrete examples and safety guidance are included.

When to use it

Load this skill when an agent must interact with or automate web pages: filling forms, logging in, scraping structured data, clicking or navigating, taking screenshots, or running multi-step browser workflows. Also useful for debugging install/connectivity issues and React introspection.

What's included

• Scripts: none bundled in the skill (has_scripts=false)
• References: extensive references and templates under references/ (has_references=true)
• Instructions: quickstart commands, the core loop (open → snapshot → click → re-snapshot), examples for reading/extracting content, waiting strategies, session and auth management, network mocking, video recording, iframe and dialog handling, and troubleshooting steps.

Compatible agents

Best with agents that can run shell tools and CLIs (Copilot/Codex/GPT CLI runners, Claude Code, Cursor). Works where a local or cloud agent can invoke agent-browser and parse snapshot JSON output.