The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around executing system commands using xcrun simctl and xcodebuild. These commands are used to boot simulators, install applications, capture screenshots, and manage device state.
Evidence includes syntax for: xcrun simctl boot, xcrun simctl install, xcrun simctl launch, and xcodebuild -project App.xcodeproj ....
Recovery steps include commands to modify the filesystem: rm -rf ~/Library/Developer/CoreSimulator/Caches.
[EXTERNAL_DOWNLOADS]: The reference documentation describes using xcodebuild -downloadPlatform iOS to fetch simulator runtimes. This is a standard operation targeting official Apple servers.
[INDIRECT_PROMPT_INJECTION]: The skill templates interpolate user-supplied data (such as <UDID>, <bundle-id>, and [args...]) directly into shell command strings. This creates a vulnerability surface where a malicious prompt could attempt command injection if the agent does not strictly validate these inputs.
Ingestion points: User-provided simulator identifiers, bundle IDs, and application launch arguments in SKILL.md and references/simctl-commands.md.
Boundary markers: The skill advises using "parsed command output or an explicit known UDID" to mitigate errors, but lacks explicit boundary markers for all interpolated fields.
Capability inventory: The skill has the capability to execute shell commands (spawn, launch), write to the filesystem (io screenshot), and delete files (erase, rm -rf).
Sanitization: The skill documentation encourages manual verification and explicit targeting, but the command templates themselves do not include sanitization logic.

ios-simulator