The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to operate through the belt CLI tool using the Bash tool. This is the primary method for interacting with the vendor's video generation models.
[EXTERNAL_DOWNLOADS]: Fetches installation instructions and documentation from the vendor's official GitHub repository (inference-sh/skills) and website (inference.sh). These are documented as trusted vendor resources.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external media content via URLs (images, video, audio) within JSON payloads passed to the generation models.
Ingestion points: Input fields such as image, video, audio, and reference_image in example belt commands.
Boundary markers: Uses structured JSON for model inputs, which helps separate data from commands.
Capability inventory: Utilizes the Bash tool to execute belt CLI commands for model inference and media processing.
Sanitization: The skill relies on the underlying belt CLI and the vendor's API to validate and sanitize provided URLs and parameters.

ai-video-generation