The Agent Skills Directory

[Indirect Prompt Injection]: The skill ingests data from connected Android devices, such as application names, process metadata, and UI tweak descriptors. This data is untrusted and could potentially be crafted to influence the agent's behavior. However, the skill treats this information as data to be displayed rather than instructions to be followed.
Ingestion points: The scripts/snapo-tweaks tool reads data from the Android device via ADB (specifically the apps, list, get, and watch commands).
Boundary markers: The skill uses structured JSON output formats and the SKILL.md instructions recommend quoting user-supplied names.
Capability inventory: The scripts/snapo-tweaks script uses subprocess.run to execute adb commands for device discovery, port forwarding, and system information retrieval.
Sanitization: The script performs type validation for incoming tweak values (e.g., checking for specific types like integer, boolean, or enum) and uses standard JSON serialization for output.
[Command Execution]: The skill includes a Python script (scripts/snapo-tweaks) that executes adb (Android Debug Bridge) commands. This is a core requirement for the skill to communicate with an Android device. The script follows safe practices by passing arguments as a list to subprocess.run, which mitigates common shell injection risks.

snap-o-tweaks