create-multi-product-hyper-motion-ad
Fail
Audited by Gen Agent Trust Hub on Jul 10, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill provides explicit instructions to extract sensitive secrets from local environment files using shell commands. In
SKILL.md, it suggests usingexport ELEVENLABS_API_KEY=$(grep '^ELEVEN_LABS_API_KEY' /path/to/gtm-goose/.env | sed 's/.*= *//')to harvest credentials. - [DATA_EXPOSURE]: The file
tests/smoke-test.mdcontains a hardcoded absolute file path referencing a specific user's desktop environment:/Users/0xhbam/Desktop/Cursor/gtm-goose/.env. This pattern is indicative of scripts designed to target specific user environments for credential harvesting. - [COMMAND_EXECUTION]: The skill relies heavily on executing shell commands and external CLIs (
higgsfield,ffmpeg,curl) with arguments dynamically constructed from potentially untrusted external data (Shopify API responses and user-supplied briefs). - [DATA_EXFILTRATION]: Extracted credentials and generated content are transmitted to external endpoints including
api.elevenlabs.ioandqueue.fal.run. While these are known services, the method of credential acquisition (direct file scraping) is unsafe. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting data from external Shopify product feeds and brand guidelines without explicit sanitization before interpolating them into prompts for AI generation models (Higgsfield and ElevenLabs).
Recommendations
- AI detected serious security threats
Audit Metadata