The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external push notifications (FCM/APNs) to handle background VoIP events, creating a potential injection surface.
Ingestion points: In SKILL.md, the messaging().setBackgroundMessageHandler callback passes remoteMessage.data directly to TelnyxVoiceApp.handleBackgroundPush for processing.
Boundary markers: No delimiters, sanitization, or validation logic is present in the provided instructions to isolate or verify the push notification payload content.
Capability inventory: The skill grants the agent capabilities to perform sensitive VoIP actions including answer(), hangup(), mute(), and sending dtmf() signals based on SDK state.
Sanitization: No sanitization of the push payload data is described before it is processed by the internal SDK components.
[CREDENTIALS_UNSAFE]: The skill documentation explicitly states that the SDK automatically persists sensitive authentication credentials in cleartext within local storage.
Sensitive data: The library manages keys such as @telnyx_password (SIP password) and @credential_token (JWT) internally.
Risk: The use of AsyncStorage for sensitive secrets is an unsafe practice as it is an unencrypted persistent store in React Native. On rooted or compromised devices, these credentials could be extracted by other malicious processes.

telnyx-webrtc-client-react-native