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
Risk Level
MEDIUM
Analyzed
Sep 23, 2026, 08:05 AM
Security Audit — agent-trust-hub — ml-engineer