The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection as it processes untrusted user-supplied content, such as speech scripts and avatar appearance prompts, which are then transmitted to an external AI service for processing.
Ingestion points: Script text inputs in references/video-generation.md and descriptive prompts for AI-generated avatars in references/photo-avatars.md.
Boundary markers: Code examples lack explicit boundary markers or instructions designed to isolate user-supplied content from API control logic.
Capability inventory: The skill possesses capabilities for network operations (interacting with HeyGen APIs) and local file system writes (saving generated assets).
Sanitization: No validation or sanitization mechanisms are included in the documented examples to mitigate the risk of malicious payload injection within user scripts.
[DYNAMIC_EXECUTION]: The integration with Remotion documented in references/remotion-integration.md involves bundling and executing code at runtime to generate video compositions, which constitutes a dynamic execution pattern.
[COMMAND_EXECUTION]: The reference documentation includes shell-based command examples (using curl) for API interactions across multiple files, defining the operational methods used by the agent to perform tasks.
[EXTERNAL_DOWNLOADS]: The skill performs downloads of video and image assets from official HeyGen infrastructure (heygen.com, heygen.ai). These references target well-known service domains essential to the skill's primary function.

avatar-video