nanogpt

Pass

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches educational training datasets from reputable platforms, including the TinyShakespeare text from GitHub and the OpenWebText dataset from HuggingFace.
  • [DYNAMIC_EXECUTION]: Employs PyTorch's torch.compile functionality to optimize the transformer's computational graph at runtime for improved execution speed.
  • [DYNAMIC_EXECUTION]: Uses the pickle module for the serialization and deserialization of model metadata, such as vocabulary mappings, during the data preparation and training processes.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for untrusted data by reading external text files (Shakespeare, OpenWebText) or user-provided datasets to train models. While these inputs are unvalidated, the impact is confined to the weights of the locally trained model, and the skill includes no automated execution pathways for the model's generated text.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 29, 2026, 03:22 PM