The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends the installation of an external agent skill (belt-sh/cli) using npx skills add and references installation guidelines from a remote GitHub repository (raw.githubusercontent.com/inference-sh).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of image and audio URLs provided as input to AI models, which could be exploited to influence model behavior.
Ingestion points: External URLs for source portraits and audio files are used as inputs for the belt command in SKILL.md.
Boundary markers: None are defined to delimit the external media data or instruct the models to ignore embedded signals.
Capability inventory: The skill utilizes the Bash tool to execute belt app run commands for model inference (transcription, lipsync, and video generation).
Sanitization: There is no evidence of validation or sanitization of the remote media content before it is processed by the AI models.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to run the belt CLI for model execution, though this is constrained by the allowed-tools configuration to only allow belt commands.

talking-head-production