qa-testing-android
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill provides CI configuration examples in
references/android-ci-optimization.mdthat usesudoto modify system udev rules at/etc/udev/rules.d/99-kvm4all.rules. This is a specific, industry-standard requirement for enabling KVM hardware acceleration for Android emulators on Linux-based CI runners like GitHub Actions. - [EXTERNAL_DOWNLOADS]: The documentation references external installers for mobile testing tools, specifically Maestro, through its official website and download subdomains.
- [REMOTE_CODE_EXECUTION]: The skill includes an installation command for Maestro using a
curl | bashpattern targetinghttps://get.maestro.mobile.dev. This is the official and well-known installation method for this service. - [COMMAND_EXECUTION]: The skill involves the execution of various shell commands via
adbandgradlefor application management, emulator control, and test execution. These are standard operations within the Android development ecosystem. - [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for agents to interact with application UIs by selecting and asserting on elements based on their text content (e.g., using
onNodeWithText). This introduces a potential surface where untrusted text rendered within a test application could influence the agent's logic. - Ingestion points: UI element text, labels, and descriptions ingested via Compose and Espresso matchers in
references/compose-testing.mdandreferences/espresso-patterns.md. - Boundary markers: None identified; the skill follows standard automated testing interaction patterns.
- Capability inventory: Employs
adb shellfor device commands andgradlefor build/test tasks. - Sanitization: There is no explicit sanitization of text extracted from the application's UI before it is used by the agent.
Audit Metadata