The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interfaces with xcrun simctl and idb for simulator management and UI automation. The script scripts/ios-sim.mjs uses node:child_process.spawn with shell: false, ensuring that arguments are passed as a discrete array which prevents shell-level command injection.
[EXTERNAL_DOWNLOADS]: The documentation references the idb toolset from Facebook's official Homebrew tap and PyPI registry. As Facebook is a trusted organization, these resources are handled neutrally in this analysis.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by reading and acting upon potentially untrusted UI data from within the simulator.
Ingestion points: UI element descriptions, labels, and titles are ingested from the simulator via idb ui describe-all in the scripts/ios-sim.mjs file.
Boundary markers: The output does not include explicit delimiters or system warnings to the agent to ignore instructions that might be embedded in UI text.
Capability inventory: The skill provides significant control over the simulator, including app installation, UI interaction (tapping/typing), and modification of system state (privacy grants, erasing/deleting devices).
Sanitization: There is no semantic validation or filtering performed on the UI text data before it is presented to the agent context.

ios-simulator