agent-browser

Automate browser workflows using accessibility tree snapshots and stable reference IDs.

1|Updated May 20, 2026
One-click install
npx skills add https://github.com/Wanli-Lee/CUA-Claw-Harness --skill agent-browser-wanli-lee
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-browser
Source: https://github.com/Wanli-Lee/CUA-Claw-Harness/tree/main/skills/agent-browser
Command: npx skills add https://github.com/Wanli-Lee/CUA-Claw-Harness --skill agent-browser-wanli-lee

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill solves unreliable, non-deterministic web automation by letting AI agents interact with pages using accessibility tree snapshots and stable reference IDs instead of brittle selectors.

Core Features & Use Cases

  • Accessibility tree snapshotting: Capture interactive elements from the rendered DOM via accessibility snapshots, returning structured JSON.
  • Ref-based deterministic interactions: Click, fill, type, hover, scroll, drag, and select using ref IDs (for example, @e2) derived from the latest snapshot.
  • Agent-friendly session control: Use isolated browser sessions plus save/load of auth state to avoid repeating login flows.
  • Operational tooling for real workflows: Wait for elements, URLs, text, or network idle; generate screenshots/PDFs; and inspect or control network requests.

Quick Start

Use agent-browser to open a URL, snapshot interactive refs, then click and fill elements by their ref IDs.

Frequently Asked Questions about agent-browser

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

FAQPage Schema
How do I automate browser workflows without brittle CSS selectors?

Browser automation can be made deterministic by using accessibility tree snapshots to select and operate UI elements via stable reference IDs instead of brittle CSS selectors.

What is an accessibility tree snapshot for web automation?

An accessibility tree snapshot captures interactive elements from the rendered DOM into structured JSON, providing AI agents with stable reference IDs for deterministic clicking and filling.

How do I fill forms and click elements in a single agent session?

To fill forms and click elements, open a URL, snapshot interactive refs, then execute click, type, and select commands using the generated ref IDs from the latest JSON snapshot.

Can I save and load authentication state for web scraping tasks?

Yes, you can save and load authentication state within isolated browser sessions to avoid repeating login flows during multi-step agent tasks and web scraping operations.

Does agent-browser support waiting for network idle or specific URL changes?

Yes, agent-browser supports waiting strategies for elements, URLs, text, or network idle, ensuring reliable execution when navigating complex single-page applications.

Why are my web automation scripts failing on dynamic single-page applications?

Dynamic SPAs often break automation due to non-deterministic UI changes; using ref-based commands driven by the latest accessibility snapshot JSON ensures stable element selection.