android-cli
Pass
Audited by Gen Agent Trust Hub on Feb 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous examples of executing system commands via the
androidCLI. These include creating/deleting virtual devices (android avd create/delete), managing SDK packages (android sdk install/uninstall), and configuring .NET workloads (android sdk/jdk dotnet-prefer). While these are legitimate functions of the tool, they grant the agent significant control over the development environment. - [EXTERNAL_DOWNLOADS]: The skill documents commands that fetch content from the internet. Specifically,
dotnet tool install -g AndroidSdk.Tooldownloads the CLI tool itself, andandroid sdk downloadfetches Android SDK command-line tools. These operations rely on the integrity of the remote repositories and network connections. - [REMOTE_CODE_EXECUTION]: The command
android device install --package <apk-path>enables the installation of third-party Android application packages (APKs) onto connected devices or emulators. Executing these applications constitutes remote code execution on the target device. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted data from external sources.
- Ingestion points: Data enters the agent's context through
android apk manifest info(reading APK files inreferences/apk-commands.md) andandroid device info(reading device properties inreferences/device-adb.md). - Boundary markers: There are no instructions or delimiters provided to warn the agent to ignore potentially malicious instructions embedded within the APK manifests or device property strings.
- Capability inventory: The agent has the capability to execute commands (
dotnet tool install), download files (android sdk download), and install/execute code on devices (android device install). - Sanitization: There is no evidence of sanitization or validation of the external data before it is presented to the agent's context.
Audit Metadata