krea-marketing
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from product URLs and design inspiration sites to inform its creative workflows, creating an attack surface for indirect prompt injection.
- Ingestion points: Processes content from user-provided URLs and design galleries such as Behance and recent.design (specified in
workflows/product-packaging-design.md). - Boundary markers: The workflows use structured labels and sections (e.g.,
SHOT 1,LAYOUT) to organize prompt generation. - Capability inventory: The skill can execute local media tools (
ffmpeg), perform network requests (curl), and use powerful generation tools (generate_video). - Sanitization: The skill includes strong defensive instructions in
workflows/product-packaging-design.md, telling the agent to treat third-party page text as untrusted evidence and to never follow instructions found within those pages. - [EXTERNAL_DOWNLOADS]: The skill downloads content or extends its capabilities through well-known external platforms.
- Audio Generation: Fetches music from the official ElevenLabs API in
workflows/launch-teaser.md. - Capability Extension: Instructs the agent to install the
hyperframesskill from a well-known vendor (HeyGen) to facilitate complex video rendering tasks.
Audit Metadata