The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze external video content and social media posts to reverse-engineer editing styles, creating a surface for indirect prompt injection.
Ingestion points: External video transcripts and metadata fetched using watch-video or social-fetch tools (referenced in references/edit-anatomy.md).
Boundary markers: The prompt instructions in SKILL.md and references/edit-anatomy.md do not include specific delimiters or instructions to ignore embedded commands within the fetched transcripts.
Capability inventory: SKILL.md guides the use of programmatic video frameworks like Hyperframes and Remotion which involve script execution and rendering.
Sanitization: There is no explicit requirement in the skill instructions to sanitize or filter external text before processing it for editing patterns.
[COMMAND_EXECUTION]: The skill recommends standard shell commands for initializing video projects, including npm install hyperframes and npx create-video@latest. These are legitimate commands for the stated purpose of the skill.

video