item-detail
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/lib/providers.mjsutilizeschild_process.spawnto execute the vendor'sdlazyCLI tool, which serves as the primary interface for triggering image generation and model-specific tasks.\n- [EXTERNAL_DOWNLOADS]: The skill performs network requests usingfetchinscripts/gen.mjsandscripts/lib/providers.mjsto download reference images from remote URLs and retrieve processed assets from AI provider APIs.\n- [DATA_EXFILTRATION]: The skill transmits local image files and prompt text to external AI service providers (including OpenAI, Google, and Replicate) and the vendor's API infrastructure (api.dlazy.com) as part of its documented generative workflow.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted product descriptions and images provided by users, which are interpolated directly into prompts for image generation models. Evidence chain: 1. Ingestion points:skill.mdprompt templates andscripts/gen.mjstask arguments; 2. Boundary markers: None (though layout delimiters are suggested for aesthetics); 3. Capability inventory:scripts/lib/providers.mjs(spawn) andscripts/gen.mjs(network and file write access); 4. Sanitization: None.
Audit Metadata