swiftui-liquid-glass
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references external resources outside the verified vendor scope, including a GitHub repository (conorluddy/LiquidGlassReference) and a private technical blog (donnywals.com) for API guidance.
- [PROMPT_INJECTION]: The skill uses authoritative language to define the capabilities and constraints of a non-existent API ("iOS 26+ Liquid Glass"). This overrides the model's baseline training with fictional technical specifications, which can lead to the generation of non-functional or invalid code.
- [PROMPT_INJECTION]: The skill workflows involve processing user-supplied code for review and refactoring, which creates a vector for indirect prompt injection where instructions within the reviewed code could manipulate the agent's behavior.
- Ingestion points: Workflow paths for "Review an existing feature" and "Improve a feature using Liquid Glass" in SKILL.md accept arbitrary code for analysis.
- Boundary markers: The instructions do not define clear delimiters or "ignore" instructions to isolate the code being processed from the agent's control context.
- Capability inventory: No sensitive file access or network tools are currently scoped in the frontmatter, limiting the potential impact of an injection attack.
- Sanitization: No input validation, escaping, or instruction filtering is implemented for the content being reviewed.
Audit Metadata