mlx-model-porting
Warn
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The benchmark runners located in 'assets/benchmarks/runners/attested_mlx_port.py' and 'assets/benchmarks/qwen2.5-0.5b/runner.py' use the 'importlib' library to dynamically load the 'model' module from a computed path in the user's cache directory. This pattern allows for the dynamic execution of generated Python code.
- [COMMAND_EXECUTION]: The skill workflow is centered around the execution of shell commands and internal Python scripts to perform model analysis and conversion tasks.
- [PROMPT_INJECTION]: Processing untrusted third-party model configuration files and documentation from local or remote repositories provides a surface for indirect prompt injection, where malicious instructions could be embedded in the model metadata and processed by the agent.
- [EXTERNAL_DOWNLOADS]: The skill contains logic and references to fetch data from well-known and trusted external services such as the GitHub API and arXiv.
Audit Metadata