The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses credential files located at ~/.heygen/credentials and .env to authenticate API requests to the HeyGen media generation platform.
[EXTERNAL_DOWNLOADS]: Automatically installs necessary Python dependencies such as google-genai, transformers, torch, and numpy from official registries to support local audio processing and generation.
[COMMAND_EXECUTION]: Spawns subprocesses to execute ffmpeg and ffprobe for audio transcoding and metadata extraction, and uses python3 to run internal generation recipes.
[EXTERNAL_DOWNLOADS]: Connects to api.heygen.com to fetch voice metadata, perform text-to-speech synthesis, and download media assets.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting external data from audio_request.json to drive AI generation.
Ingestion points: audio_request.json (prompt and query fields in audio.mjs).
Boundary markers: No explicit delimiters or instructions are used to separate untrusted user inputs from the system instructions sent to AI models.
Capability inventory: The skill has access to subprocess execution (python3), network requests (fetch), and file system operations (writeFileSync).
Sanitization: Inputs are stringified and embedded in JSON payloads or Python script templates without specialized filtering for adversarial instructions.

hyperframes-media