The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands using xcrun, simctl, and devicectl to interact with iOS simulators. Evidence:
verify-duo-rotation.ts uses execFileSync to call xcrun devicectl and xcrun simctl.
verify-duo-projection.py uses subprocess.check_output to call xcode-select.
[DYNAMIC_EXECUTION]: The Python verification script executes a renderer binary built as part of the local project. Evidence:
scripts/verify-duo-projection.py uses subprocess.Popen to launch the binary specified by the --renderer argument (defaulting to a path within the local serve-sim package).
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local simulator tools and a development server to verify device state, creating a surface for indirect prompt injection. Evidence:
Ingestion points: Output from xcrun simctl io enumerate and devicectl device orientation get in verify-duo-rotation.ts; configuration fetched from http://localhost:${port}/helper/${udid}/config.
Boundary markers: The scripts do not use specific boundary markers or "ignore" instructions when processing this data, relying on structured parsing (JSON, regex).
Capability inventory: The skill has the capability to execute shell commands and perform network operations on localhost.
Sanitization: Data is parsed as JSON or validated against specific expected strings (e.g., orientation names), but no broader sanitization is applied to the raw tool output.

serve-sim-duo-verification