foundation-models-on-device
Pass
Audited by Gen Agent Trust Hub on Apr 1, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill contains architectural patterns and Swift code snippets for building on-device AI features. All operations described are local to the device, prioritizing user privacy and offline functionality.
- [SAFE]: No suspicious network activity, hardcoded credentials, or obfuscated content were detected. The skill focuses on using official or projected system frameworks for generative AI tasks.
- [DATA_EXPOSURE_EXFILTRATION]: No sensitive file access or unauthorized data transmission patterns are present. The documentation explicitly emphasizes that no data leaves the device.
- [INDIRECT_PROMPT_INJECTION]: The skill describes patterns for processing untrusted user input via
LanguageModelSession.respond(to:)and provides a mechanism for tool calling (e.g.,RecipeSearchTool). It mitigates common risks by recommending structured output via the@Generablemacro and the use of explicit system instructions to guide model behavior and safety.
Audit Metadata