foundation-models-on-device

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The content serves as a documentation resource for on-device LLM integration and does not include any active malicious logic or scripts.
  • [PROMPT_INJECTION]: The skill demonstrates patterns for building applications that process untrusted user input, which creates a potential surface for indirect prompt injection.
  • Ingestion points: User-provided strings are passed directly to LanguageModelSession.respond(to:) and streamResponse(to:) in SKILL.md.
  • Boundary markers: The provided code examples do not illustrate the use of delimiters or specific instructions to ignore embedded commands.
  • Capability inventory: The documentation describes the use of a Tool protocol (e.g., RecipeSearchTool) which allows the model to trigger application logic.
  • Sanitization: No explicit input validation or sanitization logic is shown in the integration patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 04:01 AM
Security Audit — agent-trust-hub — foundation-models-on-device