instrumenting-with-mlflow-tracing

Pass

Audited by Gen Agent Trust Hub on Apr 23, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides documentation and code snippets for a well-known observability tool (MLflow). All external resources, such as Python and Node.js packages, refer to legitimate, widely-used libraries.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill explicitly promotes security-positive behavior by providing a 'PIIRedactionProcessor' implementation in references/advanced-patterns.md to prevent sensitive data like emails, credit card numbers, and API keys from being logged to traces.
  • [CREDENTIALS_UNSAFE]: Example configurations in references/production.md use standard environment variable placeholders (e.g., MLFLOW_TRACKING_PASSWORD="password") and correctly recommend against hardcoding secrets.
  • [EXTERNAL_DOWNLOADS]: The skill references installation of standard packages (mlflow, mlflow-tracing) from official registries (PyPI, NPM). These are documented neutrally as required setup steps.
  • [INDIRECT_PROMPT_INJECTION]: While the skill involves processing trace data that may contain LLM inputs/outputs, the provided verification code in SKILL.md only inspects structural metadata (span names and types) rather than raw content, minimizing the surface for indirect injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 23, 2026, 06:02 AM