The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to process external automation files.
Ingestion points: The skill executes maestro test in references/maestro.md, which reads external flow definitions from a provided directory or file.
Boundary markers: The instructions establish clear session boundaries through atomic leases and specific device identification, reducing the risk of cross-session interference.
Capability inventory: The skill has the capability to execute shell commands via adb, xcrun, and maestro for device mutation, app installation, and UI automation.
Sanitization: There are no specific instructions regarding the sanitization or content validation of the Maestro flows before execution.
[SAFE]: The skill uses a secure resource coordination pattern by creating atomic locks in a user-scoped temporary directory. This practice ensures mutual exclusion on shared hosts without requiring elevated privileges or exposing sensitive data.
[SAFE]: All command execution is restricted to standard development tools and is strictly bound to specific device instances via serial numbers or UDIDs to prevent accidental or malicious actions against unrelated devices.

running-mobile-simulators