appfunctions
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes
adb shellcommands (e.g.,adb shell cmd app_function) to interact with connected Android devices or emulators. This is standard functionality for Android development and testing tools. - [PROMPT_INJECTION]: The skill instructions in
references/adb-interaction-testing.mddirect the agent to treat thedescriptionfield in AppFunction metadata as a set of mandatory instructions. This creates an indirect prompt injection surface where a processed application could potentially provide malicious metadata to influence the agent. - Ingestion points: The output of
adb shell cmd app_function list-app-functionscontains thedescriptionmetadata field in JSON format. - Boundary markers: Absent; the agent is instructed to rigorously follow any instructions found within the description string.
- Capability inventory: The skill enables execution of functions on the device (
execute-app-function) and management of function states (set-enabled). - Sanitization: No sanitization is performed on the metadata before the agent is instructed to follow it.
Audit Metadata