The Agent Skills Directory

[PROMPT_INJECTION]: The skill's metadata and technical instructions contain deceptive claims regarding supported platform versions, specifically targeting 'iOS 26+' and 'Swift 6.2'. These versions do not currently exist. The skill also documents non-existent APIs, such as '.tabBarMinimizeBehavior' and specific '.presentationSizing' behaviors for iOS 26, which could mislead an AI agent into generating non-functional code.
[EXTERNAL_DOWNLOADS]: The skill contains a reference to an external documentation resource at 'https://sosumi.ai/documentation/xcode/supporting-universal-links-in-your-app'. This domain is not an official source for Apple developer documentation and is categorized as an untrusted external reference.
[PROMPT_INJECTION]: The patterns described for handling deep links and universal links establish a surface for indirect prompt injection, where untrusted external data influences the application state. 1. Ingestion points: Untrusted URL data enters the system via '.onOpenURL' and '.onContinueUserActivity' handlers in 'SKILL.md' and 'references/deeplinks.md'. 2. Boundary markers: The documentation includes a checklist item and pitfalls warning for URL validation, but no programmatic boundaries or delimiters are implemented in the provided code examples. 3. Capability inventory: The provided patterns involve programmatic navigation and sheet presentation based on parsed URL components. No high-risk capabilities like file system writes or subprocess execution are triggered by these data paths in the snippets. 4. Sanitization: The documentation advises validation but does not provide specific implementation details for sanitizing 'url.lastPathComponent' or other URL components before use in routing.

swiftui-navigation