ml-experiment-standards

Pass

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, such as machine learning notebooks and feature tables, for auditing and verification purposes. This creates an indirect prompt injection surface where malicious instructions embedded in the user-provided data could attempt to influence the agent's reasoning. The instructions do not explicitly mandate the use of boundary markers or sanitization for this content.
  • [COMMAND_EXECUTION]: The skill contains Python code templates for experiment reproducibility and spatial cross-validation. These templates include instructions for setting up environments and installing dependencies (e.g., pip freeze). While these are standard developer practices, they involve the generation of code for local execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 23, 2026, 12:19 PM
Security Audit — agent-trust-hub — ml-experiment-standards