The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from target TV applications, which could contain malicious instructions aimed at the agent.
Ingestion points: UI element labels and tree structures retrieved via the describe tool; application console logs accessed via debugger-log-registry; network logs.
Boundary markers: Absent. There are no instructions for the agent to use delimiters or treat UI/log data as untrusted.
Capability inventory: Use of tv-remote for navigation, keyboard for typing, and debugger-evaluate for code execution.
Sanitization: Absent. Data from the device is used directly without validation.
[DYNAMIC_EXECUTION]: The skill allows the execution of arbitrary code within the target device environment.
Evidence: The debugger-evaluate tool enables the execution of JavaScript strings on the Vega (Fire TV) runtime. Additionally, reinstall-app allows installing .vpkg package files, which involves loading and executing third-party binaries or assets on the target device.

argent-tv-interact