local-ai-models
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFENO_CODEEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection (SAFE): The skill provides Swift templates in
references/mlx-swift/advanced-patterns.mdthat process text inputs for structured extraction and tool calling. These code snippets interpolate text directly into prompts without boundary markers or sanitization. While this represents a vulnerability surface, it is a standard characteristic of LLM application development and is addressed within the skill's primary educational purpose. Ingestion points:StructuredGenerationService.extractStructuredInfo(from:). Boundary markers: Not present in code snippets. Capability inventory: Local model inference and tool execution. Sanitization: Not present. - Unverifiable Dependencies & Remote Code Execution (SAFE): The setup and quantization guides recommend installing standard Python packages (
mlx-lm) and adding Swift packages from the MLX project. While theml-exploreGitHub organization is not on the explicit whitelist provided, it is the official repository for Apple's MLX framework and is a community-standard resource. Downloads fromhuggingfaceare to a trusted organization. Sources:github.com/ml-explore(GitHub),huggingface.co/mlx-community(Hugging Face).
Audit Metadata