swiftui-liquid-glass
Warn
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: MEDIUMPROMPT_INJECTIONNO_CODESAFE
Full Analysis
- [PROMPT_INJECTION]: The skill utilizes highly deceptive metadata and steering instructions involving fictional technologies. The description and references focus on a non-existent 'iOS 26+' Liquid Glass API, which is a form of metadata poisoning intended to mislead the agent and users about the skill's actual capabilities.
- [PROMPT_INJECTION]: The reference file includes indirect instructions ('You should always seek guides on Liquid Glass when asked for help adopting new Apple design') designed to influence the agent's behavior. Evidence chain for indirect injection: 1. Ingestion point: references/liquid-glass.md; 2. Boundary markers: Absent; 3. Capability inventory: The skill has no code or execution capabilities; 4. Sanitization: Absent.
- [NO_CODE]: No executable scripts or binary files are included in the skill; it consists entirely of Markdown files.
- [SAFE]: References official documentation from Apple's developer website.
Audit Metadata