krea-marketing
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external product URLs and third-party web pages.\n
- Ingestion points: External data enters the context through product scraping and description fetching in
workflows/full-ad-campaign.mdandworkflows/product-packaging-design.md.\n - Boundary markers: The skill contains explicit instructions in
workflows/product-packaging-design.mdto treat third-party text as untrusted evidence and to disregard any embedded instructions or directives.\n - Capability inventory: The agent possesses the ability to execute local shell commands (
ffmpeg), generate image/video content, and perform authenticated writes to Meta Ads via an MCP connector.\n - Sanitization: The skill relies on clear instructional boundaries to prevent the agent from acting on instructions found within the processed data.\n- [COMMAND_EXECUTION]: The skill provides templates for using
ffmpegto process media assets locally.\n - Evidence:
references/video-ad-post.mdandworkflows/cinematic-product-ad.mdcontain shell command examples for video manipulation, captioning, and audio mixing intended for user-controlled or agent-generated assets.\n- [EXTERNAL_DOWNLOADS]: The skill references external services and dependencies for its core functionality.\n - Evidence:
workflows/launch-teaser.mdprovides instructions to install thehyperframesskill vianpxand utilizes the ElevenLabs API for music generation, both of which are provided by well-known vendors.
Audit Metadata