Feature: Autonomous Browser Operations

Overview

To expand PicoClaw's operational reach to the web, we are seeking to implement browser automation capabilities. This will allow the AI to interact with websites directly—navigating pages, extracting data, and performing actions like a human user. We are considering two primary paths for this: Chrome DevTools Protocol (CDP) for low-level, high-speed control, or ActionBook for a more abstracted, action-oriented approach, or webmcp for chrome.

Core Requirements

• Browser Driver Integration: Implement a lightweight browser controller (e.g., using chromedp for Go) to manage Chromium-based instances.

• Action Primitives: Support for fundamental web interactions:

• click(selector)

• type(selector, text)

• scroll(direction/amount)

• screenshot() for VLM-based visual reasoning.

• Headless Support: Ensure the browser can run in headless mode to minimize resource usage, staying true to PicoClaw’s "small and fast" philosophy.

• ActionBook Compatibility: Explore integration with ActionBook to leverage pre-defined web action sequences and improve the reliability of complex workflows.

Use Case Examples

1. Information Retrieval: "Search for the latest news on Go 1.24 and summarize the top 3 articles."
2. Task Automation: "Log in to my dashboard and export the weekly report as a PDF."
3. Dynamic Interaction: "Find the cheapest flight from Shanghai to Tokyo on this travel site."