The Agent Skills Directory

[SAFE]: The skill provides standard implementation patterns for iOS/macOS push notifications using official system frameworks such as UserNotifications and Intents.
[SAFE]: Security best practices are emphasized throughout the documentation, including the use of Keychain Sharing for storing decryption secrets and the recommendation to upload device tokens on every registration callback instead of using local caches.
[PROMPT_INJECTION]: The skill demonstrates processing external data from notification payloads (aps and userInfo dictionaries) in the AppDelegate, NotificationDelegate, and NotificationServiceExtension. While the logic provided is for legitimate developer tasks, this ingestion represents an attack surface for Indirect Prompt Injection if a malicious payload is received.
Ingestion points: Payload handling in didReceiveRemoteNotification, willPresent, didReceive, and NotificationService.didReceive across SKILL.md and reference files.
Boundary markers: None explicitly defined in the code snippets for AI context isolation of untrusted payload data.
Capability inventory: Filesystem access for temporary storage of attachments (FileManager.default.temporaryDirectory), network requests (URLSession.shared.data(from:)), and SiriKit intent donation (INInteraction.donate()).
Sanitization: Implementation relies on standard Swift dictionary casting and system-level JSON parsing; no specific sanitization of displayed text is mentioned.

push-notifications