The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for processing data from external sources, creating a potential surface for indirect prompt injection via media content or metadata.
Ingestion points: The skill instructs the agent to ingest remote media via URLs in references/image-loading-caching.md and user-selected assets from the system photo library in references/photokit-patterns.md.
Boundary markers: Code snippets include basic validation using Swift's guard statements and optional binding to check for valid data and successful decoding.
Capability inventory: The identified patterns focus on UI display, image downsampling, and media playback; no high-risk capabilities such as arbitrary command execution or sensitive file system access are present in the snippets.
Sanitization: The patterns rely on system framework decoders (like UIImage and AVPlayer) for format validation, but do not include explicit filtering for embedded prompt instructions in media metadata.
[DATA_EXFILTRATION]: The skill contains references to documentation and media resources hosted on external, non-whitelisted domains.
Evidence: The skill provides links to documentation at sosumi.ai and placeholder media assets at example.com throughout the reference files.

photokit