The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user data which presents a surface for indirect prompt injection.
Ingestion points: The skill accepts user-provided text scripts via the speech.text field and external media URLs through the inputs block in SKILL.md and references/examples.md.
Boundary markers: The 'Consent and disclosure' section provides instructions for the agent to refuse impersonation and verify ownership, acting as safety boundaries.
Capability inventory: The skill allows the agent to call external video generation models using the processed inputs.
Sanitization: The instructions mandate verifying model schemas via runware-run before invocation, but do not outline specific filtering for the input text content.
[EXTERNAL_DOWNLOADS]: The example recipes in references/examples.md reference media assets hosted on the vendor's domain (im.runware.ai). These are used as sample inputs for the video generation process and do not represent a security risk.

talking-avatar