The Agent Skills Directory

[SAFE]: The skill utilizes well-known technology services for video generation and programmatic editing, including HeyGen, Remotion, and Hyperframes. All external references target official repositories or established documentation sites.
[INDIRECT_PROMPT_INJECTION]: The skill defines a 'Reverse-Engineer a Viral Edit' workflow that involves processing external content from URLs using tools like watch-video and social-fetch. This creates a potential surface for indirect prompt injection if the source content contains hidden instructions.
Ingestion points: External data is ingested via watch-video (transcript and frame sampling) and social-fetch (social post metadata) in references/edit-anatomy.md.
Boundary markers: The instructions do not explicitly require wrapping ingested content in delimiters or using safety warnings to ignore embedded instructions.
Capability inventory: The skill primarily focuses on generating HTML/CSS or React code for video rendering. No high-privilege capabilities like arbitrary shell execution or filesystem modification are invoked in the scripts.
Sanitization: There are no explicit instructions to sanitize or validate the external data before it is processed by the agent.

video