The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a component that ingests data through multiple points: src, input-props, poster, and <track> attributes. These ingestion points could be used to deliver malicious instructions or content if controlled by an external source.\n- Ingestion points: The src, input-props, poster, and <track> attributes in SKILL.md accept external URLs or JSON data which are passed to the underlying player controller.\n- Boundary markers: The player uses an iframe with a sandbox attribute (defaulting to allow-scripts allow-same-origin) to isolate the embedded content from the main application context.\n- Capability inventory: The component supports script execution within the iframe and provides advanced control via a controller API, as well as media processing capabilities using VideoEncoder.\n- Sanitization: No explicit sanitization or validation of the ingested src URLs or input-props JSON data is described in the provided documentation, relying on standard browser isolation.

helios-player