swiftui-liquid-glass
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMPROMPT_INJECTIONNO_CODE
Full Analysis
- Metadata Poisoning (MEDIUM): The skill description and referenced documentation contain fabricated versioning ('iOS 26+') and non-existent API components. This deceptive metadata misleads the agent about valid technical standards.
- Prompt Injection (MEDIUM): The skill uses authoritative workflow instructions to force the agent to adopt hallucinated code patterns, effectively poisoning the model's internal knowledge base for SwiftUI development.
- Indirect Prompt Injection (LOW): The skill processes untrusted user input for code reviews and refactors without adequate protection against malicious code blocks. 1. Ingestion points: User-provided SwiftUI code in 'SKILL.md' workflows. 2. Boundary markers: Absent; no instructions to isolate user code from skill logic. 3. Capability inventory: High-level code review and generation (no script-based capabilities). 4. Sanitization: None identified for user-provided strings.
Audit Metadata