senior-data-scientist

Pass

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided CSV files using scripts/feature_engineering_pipeline.py and scripts/model_evaluation_suite.py. These scripts output data summaries, including raw values from the files (e.g., top categorical levels), directly into the agent's context. This creates a vulnerability where instructions embedded in the data could influence the agent's behavior.
  • Ingestion points: External CSV data files read by the load_csv() function in the Python scripts.
  • Boundary markers: Absent; the tool output is presented as unformatted plain text to the agent.
  • Capability inventory: The skill executes Python scripts locally and performs file system operations (read/write).
  • Sanitization: Absent; the scripts do not escape or validate data values before reporting them in summaries.
  • [COMMAND_EXECUTION]: The skill relies on the execution of local Python scripts to perform its core statistical tasks. While these scripts are limited to the standard library, they represent an execution capability involving interactions with the local file system for processing inputs and saving analysis reports.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 11, 2026, 05:41 PM