The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection via ingestion points like the instruction, vocabulary, and srt_key fields in the translation request metadata.
Ingestion points: Data provided by users or external files enters the agent context through parameters passed to the HeyGen API in SKILL.md.
Boundary markers: The implementation examples do not include delimiters or specific instructions for the AI to ignore instructions embedded within these fields.
Capability inventory: The skill performs network operations using fetch and the requests library to communicate with the API.
Sanitization: The code snippets do not show validation or escaping of input strings before they are sent to the external service.
[DATA_EXFILTRATION]: The skill transmits video URLs and configuration data to api.heygen.com, which is a well-known service for AI video processing. This network activity is the primary function of the skill and is documented neutrally.
[DATA_EXFILTRATION]: The API documentation includes support for a callback_url parameter, which allows the external service to transmit job completion data to a user-defined endpoint.

video-translate