The Agent Skills Directory

[SAFE]: A comprehensive review of the skill's instructions and reference implementation patterns shows no evidence of malicious code, unauthorized data access, or command injection. The skill serves its stated purpose as a developer guide for communication safety features. The referenced domains (e.g., sosumi.ai) are standard documentation placeholders for the described ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill identifies surfaces for ingesting untrusted data that is later presented to parents for approval.
Ingestion points: Data enters via CommunicationHandle values (phone/email), PersonInformation name strings, and SignificantAppUpdateTopic descriptions.
Boundary markers: The system utilizes strongly-typed objects and iMessage-based delivery to isolate the communication flow from the application context.
Capability inventory: The skill uses AskCenter.shared.ask for UI presentation and AskCenter.shared.responses for status monitoring; no arbitrary file or network capabilities are exposed to this data.
Sanitization: While not explicitly detailed, the reliance on system-level components for UI rendering provides a structural layer of protection against direct prompt manipulation within the safety prompts.

permissionkit