tao-finetune-nv-tesseract-forecasting
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation describes cloning the official source code from NVIDIA's GitHub repository and downloading pretrained model weights from Hugging Face. These operations are performed on trusted platforms and are essential for the skill's primary functionality.\n- [COMMAND_EXECUTION]: Instructions are provided for setting up the environment using 'uv' and executing training scripts ('finetune_example.py') and hyperparameter optimization trials. These commands are typical for machine learning development workflows and do not exhibit malicious patterns.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided CSV data for forecasting and training purposes. While this introduces a data ingestion surface, the risk is minimal as the model focuses on numerical time-series analysis rather than natural language prompt interpolation.\n
- Ingestion points: Input dataframes loaded from user-supplied CSV files in 'perform_forecasting' and AutoML trial scripts.\n
- Boundary markers: None explicitly mentioned for individual data points in CSV files.\n
- Capability inventory: The skill is designed to execute shell commands for model training, write outputs to the file system, and download weights over the network.\n
- Sanitization: Relies on standard pandas CSV parsing; no specific sanitization or filtering of adversarial numerical patterns is documented.
Audit Metadata