design
Warn
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The skill's root and sub-component documentation reference non-existent Apple platform versions (iOS 26+ and macOS 26+) and fictional APIs including
.glassEffect(),NSGlassEffectView,UIGlassEffect, andUIGlassContainerEffect. These claims are deceptive regarding the skill's utility and the capabilities of the actual target environment (iOS 18 / macOS 15). The inclusion of non-existent documentation links to developer.apple.com further misleads the agent into assuming these are verified system features.\n- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by defining broad triggers for user-provided UI design queries while maintaining powerful file manipulation capabilities.\n - Ingestion points: Processes untrusted user input when users ask about 'Liquid Glass', animations, or modern Apple UI effects (SKILL.md).\n
- Boundary markers: The instructions lack delimiters or warnings to ignore potentially malicious instructions embedded in user queries.\n
- Capability inventory: The skill allows the agent to use
WriteandEdittools to modify the codebase based on its internal (partially fictional) knowledge base (SKILL.md, liquid-glass/SKILL.md).\n - Sanitization: No sanitization or validation of user-provided design specifications is performed before the agent attempts to write or edit code.
Audit Metadata