Verified Agent Skill record

Hive Browser Automation

hive.browser-automation provides procedures for driving a user's already-running Chrome through hive-browser commands executed from the terminal. It covers browser lifecycle handling, profile selection, navigation and waiting, screenshot-based coordinate interaction, shadow-DOM targeting, typing, rich-text editors, and timeout recovery.

Tasks

Navigate to a specified URL with hive-browser navigate, wait for the page to load, allow a single-page application time to hydrate, and inspect the result with a page snapshot or screenshot.
Use a screenshot to locate a visible control, click it with hive-browser interact using viewport-fraction coordinates, inspect focused_element, and type into the focused editable element without a selector when shadow DOM may hide it.
For rich-text editors, click the editor before typing, then verify that the entered content took effect or that the submit button's aria-disabled state changed.
When multiple Chrome profiles are connected, inspect hive-browser status or setup, select the intended connected profile with --browser-profile, and confirm that the command result echoes the intended profile.

Inputs

A URL for navigation.
A hive-browser command and its --json option.
Viewport-fraction coordinates with both values between 0 and 1.
A CSS selector, shadow-piercing selector for page shadow-query, wait text, wait selector, or text to type.
A connected Chrome profile label when account selection matters.

Outputs

Machine-readable JSON results.
Screenshots with image data and cssWidth and cssHeight metadata.
Accessibility snapshots from page snapshot or successful state-changing interact actions.
focused_element information from click actions, including focus and coordinate details when available.
Real URLs and titles from navigation and related browser operations.

Limitations and checks

The bridge attaches to the user's existing Chrome session; there is nothing to start through the browser CLI.
Selector-based operations using document.querySelector cannot see elements inside shadow roots.
The type action's selector and wait-for-selector do not support shadow-piercing >>> syntax; page shadow-query supports it.
Coordinates are viewport fractions rather than pixels, and zoomed or transformed containers can make raw CSS-pixel bounding rectangles unsuitable for coordinate input.
A timeout can be a shared transport issue rather than a browser crash or stuck browser.
Heavy evaluate scripts can be slow, particularly when reading innerText across large pages.
The supplied material does not document a standalone installation procedure.
Confirm every hive-browser command includes --json and is run through terminal_exec.