The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data, such as video scripts, appearance descriptions, and asset URLs, which are subsequently transmitted to the HeyGen API. This introduces a vulnerability to indirect prompt injection, where specially crafted inputs could be used to manipulate the generated content or the behavior of the agent during the process.
Ingestion points: Input fields like script in references/video-generation.md, appearance in references/photo-avatars.md, and sourceUrl in references/assets.md.
Boundary markers: The documentation does not specify the use of delimiters or protective warnings for these inputs.
Capability inventory: The skill enables network communication with official vendor domains (api.heygen.com and upload.heygen.com) to send data and retrieve generated assets.
Sanitization: The provided examples do not include logic for sanitizing or validating the contents of the user-provided text strings.
[DYNAMIC_EXECUTION]: The integration guide for Remotion includes functionality to bundle and render local source code at runtime, which is a standard part of that specific workflow.
Evidence: references/remotion-integration.md demonstrates using the @remotion/bundler to process local source files (e.g., ./remotion/src/index.ts) for video composition.
[EXTERNAL_DOWNLOADS]: The documentation includes functions and instructions for downloading assets from arbitrary external URLs provided by the user to be used as backgrounds or audio inputs.
Evidence: references/assets.md describes a uploadFromUrl workflow that fetches binary data from user-supplied HTTPS URLs using the fetch API.

heygen