Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent image-generation helper with disclosed image, UI, and limited web-search use, and no evidence of persistence, credential access, destructive behavior, or hidden execution.
Install this if you want an image-generation workflow that may analyze images you provide, generate new images, open a selection UI, and occasionally search the web for visual reference details. Review generated prompts or outputs when using brand names, real people, or copyrighted movie scenes, and be aware that broad trigger words may activate the skill in some ambiguous conversations.
The trigger keywords are very broad and include common terms like 'product', 'brand', 'room', 'scene', and 'style', which can cause the skill to activate in conversations not actually requesting image generation. Unintended activation can redirect user flows, invoke tools unnecessarily, and increase exposure to downstream actions like image generation or web searches without clear user intent.
The routing logic maps ambiguous, high-frequency words such as 'city', 'room', 'scene', 'brand', and 'style' directly to execution categories without clear boundaries or disambiguation. In practice, this can misclassify ordinary discussion as a generation request, increasing the chance of unintended tool use and making prompt-routing behavior easier to manipulate through incidental wording.
The manifest frames the skill as a curated image generation assistant for producing stylized 3D images. However, the scene workflow explicitly includes using web_search to look up city landmarks, brand details, or movie-scene information, which extends behavior beyond pure image generation into external information retrieval.
No suspicious patterns detected.
