ml-engineer
Warn
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONSAFE
Full Analysis
- [DYNAMIC_EXECUTION]: The scripts 'scripts/train_sklearn.py' and 'scripts/tune_hyperparameters.py' utilize 'joblib.load()' to restore model artifacts and optimization studies. Because 'joblib' is based on Python's 'pickle' module, it can execute arbitrary code during the deserialization process. Loading a malicious model or study file could lead to a compromise of the environment.
- [INDIRECT_PROMPT_INJECTION]: The skill provides automated workflows for ingesting external data from CSV and Excel files to train models and generate performance reports. This creates a potential vulnerability surface where malicious content embedded in data files could influence the agent's analysis or subsequent recommendations.
- Ingestion points: Data is loaded from user-provided file paths in 'scripts/train_sklearn.py' using 'pandas.read_csv' and 'pandas.read_excel'.
- Boundary markers: There are no explicit delimiters or specific instructions to the agent to disregard potential instructions embedded within the tabular data.
- Capability inventory: The skill possesses file-writing capabilities ('joblib.dump'), executes model training logic, and produces summary metrics.
- Sanitization: The scripts perform standard data type conversions but do not implement sanitization or filtering to detect or remove potential injection patterns within the data content.
Audit Metadata