The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for loading and playing media from external URLs (e.g., in presentPlayer(from:url:) and SwiftUI VideoPlayer). This constitutes a potential ingestion point for untrusted data. While standard for media playback, an attacker could potentially use malicious media manifests or metadata to influence the agent's context if the content is further processed.
Ingestion points: url: URL parameters in SKILL.md functions and references/avkit-patterns.md examples.
Boundary markers: The provided snippets do not include explicit instructions or delimiters for the agent to ignore instructions embedded in media metadata.
Capability inventory: Capabilities are limited to system-standard media playback, Picture-in-Picture transitions, and AirPlay routing. There are no capabilities for file system modification, arbitrary command execution, or unauthorized network exfiltration.
Sanitization: The code snippets demonstrate standard implementation but do not include specific URL sanitization or metadata filtering logic.

avkit