The Agent Skills Directory

[SAFE]: The skill documents a legitimate pipeline for generating podcast content. It utilizes platform-specific tools (infsh app run) for image generation, text-to-speech, and video merging without attempting to bypass safety controls or execute arbitrary remote code.
[INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface for indirect prompt injection as it processes external image URLs and user-supplied text through multiple AI tools.
Ingestion points: External image URLs are accepted in the phota/train and pruna/p-video-avatar steps, and user-provided script text is ingested in the inworld/text-to-speech-2 step.
Boundary markers: There are no specific delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill uses Bash, Read, Write, and Agent tools to manage the pipeline and save intermediate state files (profile.json, output.json).
Sanitization: The skill relies on the underlying inference tools to handle content safety, with no explicit sanitization step for the user-provided script or remote image metadata. Given the media-centric nature of the workflow, the risk is minimal.

ai-podcast