The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to interact with the Android Debug Bridge (ADB) and execute local Python scripts for screen analysis.
[COMMAND_EXECUTION]: The setup.sh script utilizes sudo to install system packages (android-tools-adb) on Linux environments via standard package managers.
[EXTERNAL_DOWNLOADS]: The installation script fetches a vision model using the ollama CLI, an established service for managing local AI models.
[DATA_EXFILTRATION]: Captured device screenshots are sent to a configurable server (SENTER_URL) for visual processing. Users should ensure this endpoint is trusted, as screenshots may contain sensitive personal data or credentials visible on the device screen.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection where text displayed on the mobile device could be interpreted as instructions by the vision model and subsequently executed by the agent.
Ingestion points: Text and coordinates returned from scripts/vision_helper.py after analyzing device screenshots.
Boundary markers: Absent; the agent is instructed to use coordinates and descriptions directly without delimiters or safety context.
Capability inventory: Full ADB control including shell access, text input, and application launching via bash commands.
Sanitization: No sanitization is performed on the text extracted from the screen before it is processed by the agent's logic loop.

burner-phone