The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting UI trees and screenshots from a mobile device, creating an attack surface for untrusted data to influence the agent.
Ingestion points: The pua_ready and pua_observe tools capture UI hierarchy and visual data from the connected iPhone.
Boundary markers: No specific delimiters or instructions to ignore embedded content within the device UI are defined.
Capability inventory: The agent has access to tools for shell-level setup (pua_setup) and device interaction (pua_type_text), which could be misused if manipulated by malicious content on the phone screen.
Sanitization: There is no documented process for sanitizing UI-derived data before it enters the agent context.
[COMMAND_EXECUTION]: The skill uses local tools to manage system processes for device communication.
Evidence: pua_setup(action="start") is used to initiate USB port forwarding and testing services on the host machine.
[EXTERNAL_DOWNLOADS]: The workflow involves fetching remote components during initialization.
Evidence: The pua_setup(action="fetch") command retrieves a specific version of the PUA mobile execution service.
[DYNAMIC_EXECUTION]: The skill compiles code at runtime to deploy the automation agent to the device.
Evidence: pua_setup(action="build") triggers Xcode build processes to compile and sign the Runner target for iOS.

iphone-use-setup