browse

Controls a persistent headless Chromium browser for QA testing, interaction, and screenshot capture.

Updated Mar 23, 2026
One-click install
npx skills add https://github.com/dhnpmp-tech/project-agent --skill browse-dhnpmp-tech
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: browse
Source: https://github.com/dhnpmp-tech/project-agent/tree/main/.claude/skills/gstack/browse
Command: npx skills add https://github.com/dhnpmp-tech/project-agent --skill browse-dhnpmp-tech

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires playwright, diff, bun, and includes scripts (resource) components.

What problem does it solve? Manual QA testing and site verification are slow and hard to reproduce. This Skill gives an AI agent a persistent headless Chromium browser that can navigate URLs, interact with page elements, verify page state, and capture visual evidence at roughly 100ms per command after startup. ## Core Features & Use Cases - Page Interaction & State Assertions: Navigate, click, fill forms, upload files, handle dialogs, and assert element states (visible, enabled, checked, focused) using accessibility-tree snapshots with @e refs. - Visual Evidence & Diffing: Take annotated screenshots, capture responsive layouts at mobile/tablet/desktop viewports, diff snapshots before and after actions, and compare staging versus production pages. - Session Persistence & Handoff: Cookies, tabs, and login sessions persist between commands, with a handoff mode that opens a visible Chrome window so a human can solve CAPTCHAs or MFA before the AI resumes. - Use Case: After deploying a new checkout flow, ask the agent to open the staging site, fill the form, submit it, verify the success message appears, and attach an annotated screenshot as bug-report evidence. ## Quick Start Ask the agent to open your staging site in the browser, walk through the login flow, and take a screenshot of the result.

Frequently Asked Questions about browse

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I test a website with a headless browser from the command line?

Use the browse CLI to navigate with goto, inspect the page with snapshot or text, and interact via click and fill commands. The first call starts a persistent Chromium server in about 3 seconds, then each subsequent command runs in roughly 100ms.

How to take annotated screenshots of web page elements?

Run snapshot with the -a flag to produce an annotated screenshot with red overlay boxes and ref labels, optionally specifying an output path with -o. You can also crop screenshots to a specific element or clip region using the screenshot command.

Can a headless browser handle CAPTCHAs and multi-factor login?

Headless automation cannot solve CAPTCHAs or MFA on its own. The handoff command opens a visible Chrome window at the current page so the user completes the challenge, then resume returns control to the AI with all cookies and session state preserved.

Does the browse tool work on Windows?

Yes, but Windows uses a Node.js server bundle instead of Bun because Bun cannot reliably launch Playwright Chromium on Windows. The build script generates server-node.mjs with Bun API polyfills for compatibility.

Why do element refs stop working after navigation?

Refs like @e3 point to live element locators that become stale when the page navigates or the DOM changes. Run snapshot again after any goto or major page change to get fresh refs before interacting.