instrumenting-with-mlflow-tracing
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a vulnerability surface by instructing the agent to capture the full content of retrieved data, tool outputs, and external service responses within MLflow traces, as detailed in
references/python.md.\n - Ingestion points: Untrusted data enters the context via the
@mlflow.tracedecorator andspan.set_outputs()method inreferences/python.mdandreferences/typescript.md.\n - Boundary markers: The skill does not provide instructions for using delimiters or boundary markers to isolate captured data from instructions.\n
- Capability inventory: The skill explicitly suggests using these captured traces with follow-up diagnostic tools like the
fix-agent-issueskill, creating a path for malicious data to influence agent logic.\n - Sanitization: There are no instructions for sanitizing content against embedded prompts, although PII redaction is covered as a separate concern.\n- [COMMAND_EXECUTION]: The skill provides instructions for using the Databricks CLI (e.g.,
databricks auth login) for authentication and workspace setup, which are expected operations for this integration.\n- [EXTERNAL_DOWNLOADS]: The guide recommends installing official vendor packages includingmlflowandmlflow-tracingfrom public registries.
Audit Metadata