The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes methods for reading health data (e.g., samples, statistics, workouts) from the Apple Health store. This data constitutes an untrusted ingestion point that could be used for indirect prompt injection attacks.\n
Ingestion points: Data is fetched using HKSampleQueryDescriptor, HKStatisticsQueryDescriptor, and HKStatisticsCollectionQueryDescriptor in SKILL.md, and HKAnchoredObjectQueryDescriptor in references/healthkit-patterns.md.\n
Boundary markers: Absent. There are no instructions to use delimiters or ignore embedded instructions when processing health data.\n
Capability inventory: The skill facilitates access to sensitive health metrics (read/write), monitoring real-time workouts via HKWorkoutSession, and enabling background updates with HKObserverQuery.\n
Sanitization: Absent. The provided code snippets do not include logic for sanitizing or validating retrieved health data before it is used.\n- [EXTERNAL_DOWNLOADS]: The skill includes multiple reference links to documentation hosted on an external domain (sosumi.ai), which is not recognized as a trusted organization or well-known service. Examples include https://sosumi.ai/documentation/healthkit and several other paths under the same domain.

healthkit