The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the UiDevice.executeShellCommand method, which allows the execution of shell commands on an Android device during instrumentation tests.
Evidence: The skill provides an example: device.executeShellCommand("am start -n com.android.settings/.Settings") in SKILL.md.
Context: The skill correctly identifies this approach as @Discouraged and provides safer alternatives (e.g., using Context.startActivity with <queries> entries), aligning with official Android security best practices for routine app interactions.
[SAFE]: The skill refers to official and trusted external documentation and dependencies for Android development.
Evidence: Links to developer.android.com for UiAutomator documentation and the standard androidx.test.uiautomator:uiautomator:2.3.0 library.
[SAFE]: The skill defines a surface for processing data from the Android accessibility tree which constitutes an indirect prompt injection surface.
Ingestion points: Data enters the context via UI selectors such as By.text() and By.desc() which scrape content from the device screen (SKILL.md).
Boundary markers: No specific boundary markers or instructions to ignore embedded content are described.
Capability inventory: The skill enables interactions including click, setText, and executeShellCommand (SKILL.md).
Sanitization: No specific sanitization of the UI strings is mentioned before they are processed by the framework.
Note: This is a standard characteristic of UI automation frameworks where the application under test (or other apps on screen) could potentially present adversarial text.

cross-app-tests-with-uiautomator