The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text that serves as the script for AI avatar speech.
Ingestion points: The input_text field in references/video-generation.md and references/photo-avatars.md is populated by external user scripts.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are shown in the example code snippets, although the data is transmitted within structured JSON payloads to the service provider.
Capability inventory: The skill examples demonstrate network operations (HeyGen API) and local file reading via fs.readFileSync or Python's open().
Sanitization: No explicit sanitization or filtering of the user-provided script is documented in the integration examples.
[EXTERNAL_DOWNLOADS]: The skill documentation provides code to interact with external HeyGen API endpoints.
Evidence: Requests are directed to api.heygen.com, upload.heygen.com, and files.heygen.ai for legitimate functionality such as listing avatars, uploading media assets, and fetching generated video files.
Context: These domains belong to a well-known technology service for AI video production, and the usage pattern is consistent with standard API integration.
[SAFE]: Manual review of the code snippets and instructional content shows no signs of malicious intent or security bypasses.
Environment variables (HEYGEN_API_KEY) are handled using best practices (process.env or os.environ) rather than being hardcoded.
The automated scanner flag is likely a false positive triggered by the high density of shell commands (curl) and script examples within the documentation files.

avatar-video