What problem does it solve? Agents often need to interact with live web pages—navigating, clicking, filling forms, and reading the DOM—but lack clear rules for when and how to use native browser automation tools. This Skill provides the selection manual and core loop for the deferred browser_* tool group backed by WRY or Chromium CDP. ## Core Features & Use Cases - Structured Core Loop: Defines the canonical sequence of create, navigate, snapshot, interact, wait, and close for reliable page automation. - Selection Rules: Ten concrete rules covering snapshot timing, element refs over CSS/XPath, fill vs. type, wait vs. polling, and human takeover via browser_show for CAPTCHA or OAuth. - Honest Capability Status: Documents exactly which tools are wired (click, fill, screenshot on Linux/WebKitGTK) versus not wired (tabs, cookies, network observation), preventing false claims. - Use Case: An agent must log into a web dashboard and extract a report. It loads the browser tool group, follows the core loop, uses element refs from snapshots, and calls browser_show when a login CAPTCHA requires human takeover. ## Quick Start Ask the agent to load the browser tools and navigate to a page, take a snapshot, and fill in a login form.
