web-access

What This Skill Does

The web-access skill is the core interface for all networking capabilities within the OpenClaw agent ecosystem. It serves as a unified gateway for internet interaction, abstracting complex operations such as web search, data scraping, navigation of dynamic (JavaScript-heavy) pages, and authenticated browsing. By leveraging a Chrome DevTools Protocol (CDP) Proxy, this skill enables the agent to interact with the web as a human would—navigating, clicking buttons, filling forms, and handling cookies or sessions—ensuring it can bypass common anti-bot hurdles that plague static scrapers.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/aysun168/web-access-bak

Before initial execution, ensure your environment is configured by running the dependency check script: bash ~/.claude/skills/web-access/scripts/check-deps.sh

Ensure you have Node.js 22+ installed and Chrome remote-debugging enabled (navigate to chrome://inspect/#remote-debugging and enable "Allow remote debugging for this browser instance").

Use Cases

• Complex Research: Searching for information across multiple sources and synthesizing findings while ignoring ad-heavy or irrelevant content.
• Authenticated Tasks: Accessing private information on platforms that require login, such as your personal dashboard, internal company wikis, or subscription-based content.
• Social Media Extraction: Reliably capturing data from sites with complex dynamic rendering or strict anti-scraping measures like X (Twitter), Xiaohongshu, or Weibo.
• Automated Workflow: Navigating multi-step web forms or performing repetitive GUI interactions where programmatic API calls are unavailable.

Example Prompts

1. "Find the latest market reports on renewable energy for the last quarter and summarize the key findings for me."
2. "Go to my personal dashboard at [website URL], login, and export my last three months of transaction records to a CSV."
3. "Navigate to this Xiaohongshu post and extract all comments, then filter for those that mention product pricing concerns."

Tips & Limitations

• The Philosophy of Human Browsing: When using this skill, treat the AI as a human assistant. Don't over-plan. Start with a goal, observe the DOM structure using /eval, and react based on what you actually see on the page.
• Efficiency vs. Accuracy: Prefer Jina (r.jina.ai) for summarizing long articles or blog posts to save tokens. Use CDP only when high-fidelity interaction or state maintenance (login) is required.
• Reliability: Always prefer links found within the page (DOM href) over manually constructed URLs to ensure you carry over necessary session parameters.
• Safety: Since this skill operates via a live browser, ensure you do not use it on untrusted websites that could perform malicious UI actions. Keep the agent focused on specific, verifiable tasks.