The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, such as video transcripts and social media metadata, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill uses watch-video and social-fetch (mentioned in references/edit-anatomy.md) to pull content from external platforms. It also reads local files like .agents/product-marketing.md (in SKILL.md).
Capability inventory: The skill can execute shell commands (npm, npx), perform network operations via APIs (HeyGen, Runway), and write files (Remotion/Hyperframes rendering).
Boundary markers: The instructions do not specify using delimiters or warnings to ignore instructions embedded in the external content.
Sanitization: There is no explicit requirement for the agent to sanitize or validate the external content before processing it.
[DYNAMIC_EXECUTION]: The skill requires the agent to generate code that is executed or rendered at runtime, specifically HTML/CSS for Hyperframes and React components for Remotion.
Evidence: The SKILL.md file provides code snippets and instructions for generating programmatic video frames using templates.
[EXTERNAL_DOWNLOADS]: The skill references the installation of external packages via standard package managers.
Evidence: Mentions npm install hyperframes and npx create-video@latest for setting up production environments.
[COMMAND_EXECUTION]: The skill involves the execution of shell commands to manage dependencies and trigger video rendering processes.
Evidence: CLI commands for npm and npx are included in the setup and workflow instructions.

video