apple-on-device-ai
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides defensive security guidance, explicitly warning developers to keep untrusted user content out of system instructions to prevent prompt injection and guardrail bypass.
- [EXTERNAL_DOWNLOADS]: The skill references official and well-known industry tools such as Apple's coremltools and MLX models from Hugging Face. These are documented for legitimate development purposes and originate from trusted or well-known services.
- [COMMAND_EXECUTION]: Code snippets demonstrate the use of standard ML conversion and optimization CLI tools (e.g., pip, coremltools). These operations are restricted to the development/conversion phase and do not involve unsafe runtime command execution.
- [PROMPT_INJECTION]: The skill includes instructions on prompt design that emphasize the use of safety guardrails and behavioral prohibitions (e.g., using 'DO NOT') to ensure consistent and safe AI behavior.
Audit Metadata