The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data including reference videos (via URLs) and user-supplied text to recreate animated scenes. This creates a potential surface for indirect prompt injection where instructions could be hidden in metadata or text content.
Ingestion points: Reference video files/URLs, user-supplied photos, and user-supplied text strings (Step 1).
Boundary markers: The skill references picsart-video-rules and reference-fragment-matching for scoping and handling, but lacks explicit instructions for sanitizing or delimiting user-provided text before inclusion in the scene.
Capability inventory: The skill uses picsart_media_probe and picsart_media_video_stills for analysis (Step 2) and performs file write operations to create recreation.md (Step 3).
Sanitization: No specific sanitization, filtering, or escaping mechanisms are described for the text provided by the user.

recreate-video