The Agent Skills Directory

Command Execution: The skill orchestrates UI automation by invoking the winapp CLI tool. It handles tasks such as application launching, UI tree inspection, and input simulation (keyboard, mouse, touch, pen). The provided PowerShell template uses array-based argument passing (@Arguments), which is a recommended practice to help prevent command injection by ensuring that parameters are treated as data rather than executable code.
Security Isolation (Windows Sandbox): A significant portion of the skill is dedicated to supporting Windows Sandbox (--on sandbox). This approach is a positive security pattern, as it allows synthetic UI inputs and test executions to be isolated from the user's primary desktop environment, reducing the risk of accidental interference or data exposure.
Dynamic Execution Patterns: The testing template includes the use of PowerShell's [scriptblock]::Create to dynamically define and execute logic for verifying application state, such as checking persistence files in the guest environment. This mechanism is used to facilitate complex assertions that go beyond simple CLI property reads.
Indirect Prompt Injection Surface: The skill ingests data from external sources, specifically the UI Automation tree of the applications being tested (e.g., control names, automation IDs, and property values). While this is a common surface for indirect prompt injection, the skill manages this risk by using structured CLI commands and JSON parsing to process the retrieved data rather than interpolating it directly into natural language instructions.
File and Artifact Management: The skill manages the lifecycle of test evidence, such as screenshots and video recordings, by delivering them from the guest (Sandbox) to the host environment. These operations use specific CLI verbs (push, pull, screenshot, record) with defined output paths, ensuring that file transfers are controlled and observable.

winui-ui-testing