The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted user data to generate video and audio content, creating a surface for indirect instructions.
Ingestion points: User-provided script text for inworld/text-to-speech-2, training images for phota/train, and video_prompt for pruna/p-video-avatar.
Boundary markers: None; external content is interpolated directly into JSON payloads for shell commands.
Capability inventory: The skill uses Bash to execute model inference commands and Write to save configuration and results.
Sanitization: No validation or escaping is applied to the input text or image URLs before they are passed to the external AI models.
[SAFE]: The skill integrates with established external services including Google Gemini for image generation and Inworld for text-to-speech, which is appropriate for its primary purpose of media production.

ai-podcast