skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/risk-assessment-ml/Gen Agent Trust Hub
risk-assessment-ml
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The ConstructionRiskAssessor class uses joblib.load() to deserialize model files. This pattern is functionally necessary for machine learning but can execute arbitrary code if the loaded file is malicious.
- [DATA_EXFILTRATION]: The skill requests filesystem permissions to read project history data and write risk reports in Excel format. This involves local file access for data processing, specifically reading project_history.csv and writing output via openpyxl.
- [PROMPT_INJECTION]: The skill processes untrusted external data which creates an indirect prompt injection surface. Ingestion points: project_history.csv and user-provided Excel or JSON files. Boundary markers: No delimiters or warnings are used to isolate data from instructions. Capability inventory: The skill has filesystem read and write permissions. Sanitization: No specific filtering or validation of data content is performed before the data is processed by the AI.
Audit Metadata