The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The README.md file provides installation instructions that download and execute scripts directly from a remote source.
Evidence: The command /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/QwenLM/qwen-code/main/packages/cua-driver/scripts/install.sh)" is suggested for macOS and Linux.
Evidence: The command irm https://raw.githubusercontent.com/QwenLM/qwen-code/main/packages/cua-driver/scripts/install.ps1 | iex is suggested for Windows.
[COMMAND_EXECUTION]: The skill operates by instructing the agent to invoke the qwen-cua-driver CLI binary to perform various UI automation tasks.
Evidence: Instructions frequently use shell commands such as qwen-cua-driver launch_app, qwen-cua-driver click, and qwen-cua-driver get_window_state to interact with the host operating system.
[INDIRECT_PROMPT_INJECTION]: The skill consumes untrusted data from application accessibility trees and web page content, which could potentially contain malicious instructions.
Evidence: The skill explicitly acknowledges this risk in SKILL.md and BROWSER.md, instructing the agent to treat UI labels, page text, and attributes as untrusted content that should not override system instructions or grant unauthorized approvals.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch updates and configuration from external repositories managed by the author.
Evidence: The qwen-cua-driver skills install --from main command and references to the QwenLM/qwen-code GitHub repository for installation and updates.

cua-driver