ml4t-case-study-development
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill defines a best-practice research workflow for developing trading strategies. It focuses on reproducibility, artifact management, and diagnostic gates.
- [INDIRECT_PROMPT_INJECTION]: The workflow describes the ingestion of external data files (e.g.,
.parquetand.yamlconfiguration files). - Ingestion points: The pipeline reads universe and configuration data from
setup.yaml,features.yaml, andbacktest.yaml, and financial data frometf_data.parquet(SKILL.md). - Boundary markers: None explicitly defined in the workflow description.
- Capability inventory: The workflow utilizes internal vendor library functions (
ml4t.backtest.run_backtest,ml4t.engineer.compute_features) to perform data processing and financial simulations. - Sanitization: Not explicitly mentioned, as the skill is a high-level process guide.
- Risk Assessment: While data ingestion is a theoretical surface for indirect prompt injection, the use of structured data formats and domain-specific financial libraries under the vendor's own namespace minimizes risk. The skill is assessed as safe.
Audit Metadata