The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from user-provided text and external URLs, which represents a potential attack surface for indirect prompt injection.
Ingestion points: The skill ingests voice_script text and media URLs (image, audio) for various AI model parameters in SKILL.md.
Boundary markers: Input data is structured as JSON, but there are no explicit instructions or delimiters used to separate user data from agent instructions or to ignore potential commands embedded in the inputs.
Capability inventory: The skill uses the Bash tool (restricted to the belt CLI) to perform network operations and write output files.
Sanitization: No input validation or sanitization of the external content is implemented in the documented workflows.
[EXTERNAL_DOWNLOADS]: Fetches configuration and installation guidelines from the vendor's official GitHub repository (inference-sh/skills).
[COMMAND_EXECUTION]: Utilizes the belt CLI via the Bash tool to execute AI models. The skill's tool access is restricted specifically to the belt command scope.

ai-avatar-video