ml-pipeline-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is purely documentation-based and does not include any executable scripts, binaries, or active command-line instructions. All code and YAML snippets provided are conceptual templates for workflow orchestration.
- [INDIRECT_PROMPT_INJECTION]: The skill describes an attack surface by instructing the agent to ingest and process raw data from external sources. However, it incorporates security best practices by explicitly defining 'Evaluation boundaries,' mandating data validation and quality checks (e.g., Great Expectations, TFX), and requiring versioning of data and code to ensure integrity.
- [REMOTE_CODE_EXECUTION]: While the skill references well-known MLOps tools (such as Apache Airflow, Kubeflow, and MLflow) and cloud services (AWS SageMaker, Google Vertex AI), it does not perform any remote downloads or execution. It specifically instructs the agent not to fabricate missing scripts or treat unrelated files as equivalent, serving as a safeguard against accidental execution of unverified content.
- [COMMAND_EXECUTION]: There are no shell commands or system calls present in the skill instructions. The skill utilizes standard markdown for documentation purposes without invoking any subprocesses or dynamic context injection patterns.
Audit Metadata