The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media data (videos, images, and audio) through tools like upload_asset and model_run, which presents an attack surface for indirect prompt injection where malicious instructions could be embedded in the media metadata or content.
Ingestion points: Asset uploads via upload_asset (SKILL.md).
Boundary markers: No specific boundary markers or instructions to ignore embedded content are present in the prompt templates.
Capability inventory: The skill has the capability to run AI models (model_run) and download files to the local system (asset_download).
Sanitization: No explicit sanitization of media content or metadata is performed before processing.
[COMMAND_EXECUTION]: The skill includes instructions to install additional modules using npx skills add scenario-labs/skills. This is an administrative instruction for the user to add official vendor resources from the author's repository.
[EXTERNAL_DOWNLOADS]: The documentation describes using curl -L to save generated video files from URLs provided by the platform's asset_download tool. This is the intended functional behavior for retrieving media results from the vendor's infrastructure.

scenario-video