ghost-browser_skill

Overview

This skill provides an automated Chrome browser interface driven by nodriver and optimized for LLM-controlled web tasks. It exposes a compact CLI with LLM-friendly commands for navigation, interaction, session management, screenshots, and clean markdown page reads. Commands prioritize minimal token usage and reliable element matching by visible text.

How this skill works

The skill runs a local browser daemon that accepts concise CLI commands to navigate pages, list interactive elements, interact by visible text, and produce readable markdown. It saves and restores full session state (cookies, localStorage, sessionStorage) and exposes JSON output for machine use. Fallback CSS-selector actions and debugging tools (network/console logs) are available when text-based interaction is insufficient.

When to use it

• Automated login flows and session persistence across runs
• Scraping or reading pages as clean markdown rather than raw HTML
• Form filling, file uploads, and click/hover interactions by visible labels
• Taking screenshots or generating PDFs of pages reliably
• Debugging dynamic pages with network and console logs

Best practices

• Always follow the recommended workflow: navigate → wait-ready → page-summary → elements → interact/readable
• Prefer text-based commands (interact/click/type/fill-form) before using CSS selectors
• Use readable or readable --max-length for token-efficient page extraction; avoid raw content (HTML) unless necessary
• Save authenticated state with session save and reuse via session load or profiles for persistent automation
• Run the daemon headless for CI or with named profiles for persistent local development

Example use cases

• Automate site login, save session, and perform authenticated scraping of dashboards
• Run an agent to walk a multi-step signup flow and capture screenshots at key steps
• Extract article content as markdown for summarization or ingestion into downstream tools
• Test UI flows by listing elements, clicking by label, and validating page-summary outputs
• Collect network logs and console errors for diagnosing intermittent client-side failures