The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes an application flow that integrates network-delivered data through specific callback mechanisms, creating a potential attack surface for indirect prompt injection.- Ingestion points: The skill body in SKILL.md identifies on_message, on_input, and on_unicode as callbacks for processing data from WebRTC, native, and SHM transports.- Boundary markers: The high-level lifecycle documentation does not include specific boundary markers or instructions to ignore embedded commands within the network-ingested data.- Capability inventory: The skill defines capabilities for video streaming (stream_video), NAT traversal configuration (set_webrtc_ice_servers), and server lifecycle management (start, stop, close) within SKILL.md.- Sanitization: The provided code snippets and documentation do not detail sanitization or validation routines for data received via the registered input callbacks.

application-flow