foundation-models-on-device
Pass
Audited by Gen Agent Trust Hub on Mar 26, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: No malicious patterns, data exfiltration attempts, or unauthorized system access were found in the skill content. The focus is on on-device privacy and local execution.- [NO_CODE]: The skill consists solely of a markdown file providing educational content and code snippets. There are no executable scripts (Python, JavaScript, etc.) included in the package.- [PROMPT_INJECTION]: The skill describes an architecture for processing untrusted user input via LLM tool-calling interfaces.
- Ingestion points: User prompts are passed directly to
session.respond(to:)as seen inSKILL.md. - Boundary markers: The documentation recommends using the
instructionsparameter to set system-level constraints and safety measures. - Capability inventory: The skill demonstrates custom tool invocation (e.g.,
RecipeSearchTool) which executes Swift code based on model output. - Sanitization: No explicit input sanitization or output validation logic is included in the tutorial snippets.
Audit Metadata