agent-device
Enables AI agents to automate testing and control multi-platform applications, validating the effects of code changes in real-time.
Install & Use
Copy this prompt and send it to your AI assistant (Claude / Cursor / TRAE / Codex / WorkBuddy etc.) to auto-install:
Help me install this AI Skill: agent-device. It is used for: Enables AI agents to automate testing and control multi-platform applications, validating the effects of code changes in real-time. Full Skill content: https://321skill.com/skills/agent-device/raw/index.md Read that page and install it.
The prompt includes a link to the full Skill content. You can also view the full content.
agent-device addresses the issue where AI coding agents cannot directly verify how their code modifications affect an application's behavior on real devices (e.g., phones, TVs, desktops). It provides a command-line tool that allows agents to read application accessibility snapshots, locate UI elements, perform actions like clicks and text input, and interact with the app in real-time, forming a "check-act-verify" feedback loop.
Usage is straightforward: developers or AI agents initiate a session with a target device (e.g., iOS Simulator, Android device, web browser) via the command line, then send a series of instructions (e.g., agent-device snapshot to view the interface, agent-device press @e2 to click a button) to drive the application and observe the resulting UI changes. The entire process supports screenshots and screen recordings for evidence retention.
It is particularly well-suited for developers requiring automated testing, UI validation, or AI-assisted complex application workflows, especially front-end developers, QA engineers, and agent developers. They can leverage this tool to have AI agents operate applications like real users, verifying if features work as intended.
It is recommended to run agent-device doctor first to check the environment configuration. Note that element reference identifiers (e.g., @e2) are only valid within the most recent command output; they must be re-fetched after UI changes. For applications with poor accessibility information, rely more on screenshots and video for verification.
Key Features
Unlike similar browser automation tools (e.g., Playwright), which are primarily designed for scripted testing, agent-device is specifically built for AI agents. It provides token-efficient accessibility snapshots and a concise command-line interface, making it easier for AI to understand and make decisions. Furthermore, it natively supports unified control of applications across multiple platforms including iOS, Android, tvOS, Web, and more, not limited to just browsers.
Limitations
Automation effectiveness may be suboptimal for applications that rely entirely on image recognition for interaction or for very old applications with extremely poor accessibility information.
FAQ
What is the relationship between agent-device and agent-browser?
agent-device is inspired by Vercel's agent-browser and extends its "check-act-verify" workflow to mobile and TV platforms. For the Web platform, it actually calls agent-browser to perform the actions.
Which devices and platforms are supported?
It supports applications on iOS, Android, tvOS, Android TV, Amazon Vega OS TV (via VVD virtual machine), Web, macOS, and Linux platforms.
Installation guide for AI assistants
If your AI coding assistant (Claude Code, Cursor, TRAE etc.) can see this page, send it this message to auto-install:
Visit https://321skill.com/skills/agent-device/raw/index.md to read the original Skill definition (Markdown format) for agent-device, and install it according to the instructions.
Raw Markdown URL for AI: /skills/agent-device/raw/index.md