skills.rest
gmickel
gmickelCOMMUNITY
Explore all skills by this author →
Website𝕏@gmickelGitHub

flow-next-drive

Automates UI driving across web, Electron, and native desktop apps via a driver ladder.

692|55|Updated Dec 26, 2025
Search Tags:#electron#ui automation#computer use#screenshot#Browser Testing#CDP#E2E Testing
One-click install
npx skills add https://github.com/gmickel/gmickel-claude-marketplace --skill flow-next-drive
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: flow-next-drive
Source: https://github.com/gmickel/gmickel-claude-marketplace/tree/main/plugins/flow-next/codex/skills/flow-next-drive
Command: npx skills add https://github.com/gmickel/gmickel-claude-marketplace --skill flow-next-drive

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires agent-browser, chrome-devtools-mcp, cua-driver, and includes references (resource) components.

What problem does it solve?

Verifying that a UI actually works requires driving it like a real user, but no single automation tool covers web apps, Chromium-backed desktop apps (Electron/WebView2), and genuinely native apps. This Skill detects the surface type, picks the best available driver, and degrades gracefully when richer drivers are absent.

Core Features & Use Cases

  • Surface-aware routing: Classifies targets as web apps, Chromium-backed desktop apps (driven over CDP), or true-native apps (driven via Cua Driver or Computer Use), then applies a shared observe → snapshot → act → verify → capture → release flow.
  • Driver ladder with graceful degradation: Probes agent-browser (default), chrome-devtools-mcp, Playwright, cursor-ide-browser, and native drivers top-down, falling back to manual or documented-limitation outcomes instead of failing.
  • Evidence-based verification: Every verify step checks expected UI state, a clean console, and no failed network requests, capturing screenshots and logs for downstream QA verdicts.
  • Use Case: Ask the agent to test a deployed sign-in flow: it opens the site with agent-browser, fills the form on fresh element refs, confirms the dashboard loads with no console errors, and saves a screenshot as evidence.

Quick Start

Ask the agent to check the page at your app's URL and verify the login flow works end to end.

Frequently Asked Questions about flow-next-drive

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

FAQPage Schema
How do I automate browser testing of a web app with an AI agent?▼

The skill drives web apps through the agent-browser CLI by default: open the URL, snapshot interactive elements, act on fresh refs, and verify state. It re-snapshots after every DOM change so stale element references do not cause false failures.

Can I automate an Electron or WebView2 desktop app?▼

Yes. Electron and Windows WebView2 apps are Chromium-based, so they are driven over the Chrome DevTools Protocol by launching the app with a remote debugging port and attaching agent-browser or chrome-devtools-mcp to it. They are not routed to Computer Use.

What tools drive native macOS apps that expose no CDP port?▼

True-native apps such as AppKit/SwiftUI or macOS WKWebView are driven by the Cua Driver MCP server using the accessibility tree, falling back to Computer Use when Cua is absent. On headless CI hosts, the Cua Sandbox provisions an isolated VM instead.

Does this work in headless CI environments without a display?▼

Web and Chromium surfaces work headlessly through agent-browser. Native-app driving requires a real display, so on CI the only option is the Cua Sandbox, which drives inside a provisioned VM or container; otherwise the limitation is documented rather than failing.

Why does automation report element not found or pointer-events errors?▼

Element refs go stale after any navigation, click, or form submit. The skill mandates re-snapshotting after each DOM change before acting, and treats such errors as stale snapshots until a fresh snapshot confirms otherwise.

What are the limitations of this UI automation approach?▼

iOS and iPadOS app driving is explicitly out of scope and deferred to community simulator skills. Shell-level integration like system trays and OS dialogs cannot be reached over CDP, and no driver above agent-browser is ever a hard dependency.

Related Skills

Explore More Software Engineering Skills→
VrooliVrooli
54

Scenario to Desktop

Convert Vrooli scenarios into cross-platform desktop apps via the scenario-to-desktop CLI.

Official
Advanced
joelklabojoelklabo

sketch-devtools-qa

Diagnose Chrome DevTools errors and UX regressions in the upload-to-convert flow.

Community
Intermediate
Obsidian-OwlObsidian-Owl
4

speckit-test-review

Assess test quality across unit, integration, contract, and e2e tests.

Official
Advanced
lushly-devlushly-dev
4

write-tests

Guide unit, integration, and E2E test authoring with TDD workflows.

Official
Intermediate