varg-character-keynote
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes a user-defined text string (
LINE) which is interpolated into a video generation prompt for the Seedance model. This creates a surface where malicious input could influence the generation parameters. - Ingestion points: The
LINEvariable and thecharacter/sceneprompt placeholders in the TSX template withinSKILL.md. - Boundary markers: The template uses standard string interpolation (e.g.,
"Dialogue: \"" + LINE + "\"") but does not include explicit instructions to the AI model to disregard potentially malicious instructions embedded in that input. - Capability inventory: The skill uses the
Bashtool to executecurlandjqfor sending data to thevarg.aiAPI and polling job status. - Sanitization: There are no provided validation rules or sanitization steps to filter the content of the
LINEstring before it is processed.
Audit Metadata