Browser Automation

Control a dedicated Chrome browser via agent-browser from the terminal. Token-efficient: semantic snapshots expose interactive elements as stable refs (@eN) instead of verbose accessibility trees or fragile CSS selectors.

Purpose

Operate a real browser session: navigate, click, fill forms, extract data, and call internal site APIs using the browser's cookies. Covers generic browser operations plus site-specific guides (Gmail, LinkedIn, Teams, Jira, Teamtailor, Humand.co) loaded on demand.

Use for: automating web apps that require login, scraping authenticated sites, filling forms, extracting data, navigating SPAs, job search automation, inbox management.

Do NOT use for: desktop apps, mobile emulators, API-only integrations without a browser, or captcha solving (defer to the user).

Command hierarchy (semantic-first)

Choose commands in this order. Each level down is a fallback, not a preference: