The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for processing push notification payloads from external messaging services, which constitutes a standard attack surface for untrusted data.
Ingestion points: Untrusted data enters the application via RemoteMessage.data in the MyFirebaseMessagingService class (Android) and the userInfo dictionary in the NotificationManager class (iOS).
Boundary markers: The provided implementation examples do not include explicit boundary markers or instructions to the agent to ignore potentially malicious embedded content within the payload strings.
Capability inventory: The ingested data is used to display local notifications (NotificationManager.notify, UNUserNotificationCenter) and drive application navigation through deep-link routing (DeepLinkRouter).
Sanitization: The code demonstrates basic data extraction and casting but does not include specific validation or sanitization of the notification strings before they are processed or displayed.
[EXTERNAL_DOWNLOADS]: The skill documentation includes examples of fetching content from remote sources.
The iOS NotificationServiceExtension sample demonstrates downloading media attachments from a user-provided image_url using URLSession.shared.downloadTask.
The testing documentation provides a curl command that interacts with the official Google FCM API (fcm.googleapis.com), which is a well-known technology service.

push-notifications