The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download the CLI tool from the npm registry and provides instructions to fetch skill metadata directly from Vercel Labs' official GitHub repository (raw.githubusercontent.com/vercel-labs/agent-browser).
[REMOTE_CODE_EXECUTION]: The installation process involves running agent-browser install, which downloads and executes platform-specific browser binaries (Chromium, Firefox, WebKit) required for the tool to function.
[PRIVILEGE_ESCALATION]: The documentation includes commands for installing system-level dependencies on Linux using sudo apt-get or sudo dnf, which is a standard procedure for configuring browser automation environments.
[DYNAMIC_EXECUTION]: The skill includes an eval command that allows the execution of arbitrary JavaScript within the browser context. Furthermore, integration templates for Python and Node.js use subprocess.run(shell=True) and child_process.exec, which are powerful capabilities necessary for CLI interaction but require careful input sanitization.
[COMMAND_EXECUTION]: The primary function of this skill is to provide a comprehensive interface for the agent-browser CLI, enabling actions such as navigating URLs, interacting with page elements, and managing browser sessions.
[INDIRECT_PROMPT_INJECTION]: As the tool is designed to ingest and process content from arbitrary web pages (accessibility trees and snapshots), it has an inherent attack surface for indirect prompt injection, where malicious content on a website could attempt to influence the agent's logic during automation tasks.

agent-browser