The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses AppleScript (osascript) and shell commands (screencapture) to interact with the macOS UI and capture screen data. This includes recurring screen capture loops and potential UI automation that bypasses normal application constraints.
[INDIRECT_PROMPT_INJECTION]: The workflow relies on vision_analyze to process screenshots. If the Find My app displays attacker-controlled content (e.g., a device named with malicious instructions), it could influence the agent's behavior during analysis.
Ingestion points: Screen captures of the Find My application (/tmp/findmy.png).
Boundary markers: None; the captured image is passed directly to the vision model.
Capability inventory: Execution of shell scripts, AppleScript, and network-based vision analysis.
Sanitization: None; raw screen data is processed by an LLM vision tool.
[PRIVACY_RISK]: The skill facilitates the tracking of physical locations of people and items. While intended for personal use, the automation could be used for unauthorized surveillance. Additionally, screen captures (screencapture) are stored in the public /tmp directory, where other local users or processes could access them.

findmy