neo-stealth-browser_skill

Overview

This skill provides an automated Chrome browser driven by nodriver for AI agent web tasks. It exposes a compact CLI with LLM-optimized, token-efficient commands for navigation, interaction, reading pages as markdown, session management, screenshots, and debugging. The design emphasizes visible-text interactions and session persistence to make web automation reliable and low-cost.

How this skill works

Commands control a headless or visible Chrome instance via a local daemon bound to localhost. The workflow focuses on lightweight inspection commands (page-summary, elements) and text-based interactions (interact, fill-form) to avoid costly raw HTML. Sessions, profiles, cookies, storage and logs are kept under a local state directory so authenticated contexts can be saved and restored.

When to use it

• Automate login flows and preserve authenticated sessions across runs
• Scrape or read page content as clean markdown without downloading raw HTML
• Fill forms, click buttons, and upload files using visible-text matching
• Take screenshots or export PDFs of pages for reporting or debugging
• Debug page issues with network and console logs while running automated tasks

Best practices

• Always follow the recommended workflow: navigate → wait-ready → page-summary → elements → interact/readable
• Prefer page-summary, elements, and readable over raw content to save tokens and improve accuracy
• Use text-based interact/type commands first; fall back to CSS selectors only when text matching fails
• Save and load sessions for repeatable authenticated automation instead of re-entering credentials
• Run the daemon with a named profile for persistent cookies, storage, and extensions when long-lived state is required

Example use cases

• Automated login and session save for periodic data extraction from dashboards
• Site health checks that capture screenshots, network logs, and console errors
• Form-filling bots for batch submissions or testing signup flows
• Site scraping that returns clean markdown for downstream LLM processing
• Manual QA workflows that reproduce issues with saved profiles and recorded network logs