The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references installation instructions hosted on GitHub for the inference.sh platform. These references target a well-known service and are documented for user setup purposes.
[COMMAND_EXECUTION]: The skill provides examples for using the belt CLI (inference.sh command-line tool) to execute video generation and editing tasks via belt app run.
[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface by ingesting untrusted data through parameters such as prompt, first_frame, reference_images, and video URLs.
Ingestion points: Untrusted data enters the agent context through CLI arguments passed to belt app run in SKILL.md examples.
Boundary markers: No explicit boundary markers or safety instructions are present in the examples to delimit user input from command structure.
Capability inventory: The skill utilizes the Bash tool to execute the belt CLI, which performs network operations to the inference.sh API.
Sanitization: Input sanitization is not explicitly defined within the skill documentation, relying on the underlying belt CLI and the Alibaba API for safety filtering.

happyhorse