Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent image-generation helper with some broad activation and privacy-notice caveats, but no evidence of hidden, persistent, destructive, or deceptive behavior.
Install only if you are comfortable using an image-generation skill that may send prompts, uploaded reference images, and derived search queries to image analysis, generation, or search tools. Avoid uploading sensitive personal photos, confidential brand assets, or private project details unless that processing is acceptable.
The manifest description includes many broad everyday trigger keywords such as 'room', 'product', 'style', 'brand', and 'portrait', which can cause the skill to activate in contexts far beyond the user's intent. Over-broad activation increases the chance of accidental invocation and unintended routing of user prompts or images into this skill's toolchain.
The routing logic maps very common nouns like 'city', 'room', 'scene', 'brand', and 'product' directly to generation categories without sufficient disambiguation. This can misclassify ordinary conversation as generation requests, leading to unintended tool use, unnecessary data handling, or surprising outputs.
The workflow instructs the system to send reference images to images_understand and prompts/images to gen_images, but the skill does not disclose to users that their uploaded content may be transmitted to external processing tools. This creates a privacy and consent risk, especially when users provide personal photos, branded assets, or sensitive visual content.
The skill uses web_search for city landmarks, brand details, and movie scenes, but does not warn users that their supplied topics may be sent to an external search service. While lower severity than image sharing, it still exposes user-provided interests or project details without explicit notice.
No suspicious patterns detected.
