senior-data-scientist

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill is designed to ingest and process external data and configurations which may originate from untrusted sources.
  • Ingestion points: Reads from the data/ directory and config.yaml file via script arguments.
  • Boundary markers: None identified in the provided file to separate instructions from data.
  • Capability inventory: Executes arbitrary Python scripts (python3), builds Docker images (docker build), and manages infrastructure (kubectl apply, helm upgrade).
  • Sanitization: No validation or sanitization of input data or configuration values is documented.
  • [Command Execution] (HIGH): Provides tools that execute system-level commands for model deployment and infrastructure management, creating a high-impact outcome if the agent is manipulated via injection.
  • [Unverifiable Dependencies] (LOW): References a large tech stack (PyTorch, TensorFlow, LangChain, etc.). While installation commands are not in this file, the runtime environment likely relies on these external packages.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 17, 2026, 12:23 AM