ios-26-platform

Warn

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: MEDIUMSAFE
Full Analysis
  • Metadata Poisoning (MEDIUM): The skill identifies itself as a guide for 'iOS 26 Platform' and describes fictional features like 'Liquid Glass' and 'macOS Tahoe'. This is misleading metadata that can deceive an AI agent into assuming non-existent capabilities and generating non-functional code.
  • Indirect Prompt Injection (LOW): The skill provides instructions for building components that ingest external data, such as 'WebView' and 'WebPage' (referenced in 'references/swiftui-apis.md'). This establishes a surface for indirect prompt injection where external web content could influence the agent's behavior.
  • Ingestion points: The skill provides documentation for WebView and WebPage APIs which load external URLs.
  • Boundary markers: Boundary markers and system instructions to ignore embedded content are absent in the provided code snippets.
  • Capability inventory: Includes network-enabled UI components capable of loading arbitrary URLs and processing user-directed web navigation.
  • Sanitization: The documentation lacks any mention of URL validation, content sanitization, or standard security protocols like Content Security Policy (CSP).
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 15, 2026, 10:31 PM