Security audit
Security checks for vulnerabilities and agentic risk
This is a Markdown-only prompt-writing skill with minor routing and language caveats but no sensitive access, execution, persistence, or deceptive behavior.
Before installing, consider whether you want generic image-prompt requests to use this Banana/Imagen-specific format and whether English prompts with Chinese explanations match your workflow. No sensitive permissions or automatic actions are evident from the artifact.
Mandating English prompt output without user choice can override user language preferences and create a mismatch between requested and delivered content. While not directly enabling code execution or data theft, it can degrade usability and cause unintended disclosure or processing of content in a different language context than the user expected.
The activation phrase list includes very broad requests such as generic image-prompt help, which can cause this skill to trigger outside Banana-specific contexts. That creates unintended routing and scope capture risk, where users asking for general image assistance may be forced into this skill’s assumptions and output format.
The trigger conditions do not clearly separate Banana-specific use from general image-generation tasks, making invocation ambiguous. In multi-skill environments, this can cause misrouting, override more appropriate skills, and produce outputs inconsistent with user intent or platform constraints.
The output specification requires an English prompt as a fixed deliverable and does not offer locale selection, which hard-codes behavior that may conflict with user needs. In agent settings, inflexible output rules can cause inappropriate responses, reduce accessibility, and make the skill behave unexpectedly across regions.
This file is dominated by Chinese-language instructional text and examples, which can amount to a language/locale constraint in the skill content. The policy requires either user choice or a clearly documented justification for such constraints, and no opt-in or scope limitation is stated here.
No suspicious patterns detected.
