skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/ifc-qto-extraction/Gen Agent Trust Hub
ifc-qto-extraction
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes external binaries including RvtExporter.exe, IfcExporter.exe, DwgExporter.exe, and DgnExporter.exe using the Python subprocess.run module and n8n Execute Command nodes. While the Python implementation uses a list of arguments to mitigate shell-level injection, the n8n example uses string interpolation for model paths which could be vulnerable to argument injection if paths are not properly sanitized.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external BIM models (.ifc, .rvt) and incorporates extracted metadata into reports.
- Ingestion points: Model data is ingested via load_bim_data (Excel) and the IfcOpenShell library.
- Boundary markers: No specific delimiters or instructions are provided to the agent to ignore instructions that might be embedded in model metadata (e.g., in Category or Type Name fields).
- Capability inventory: The skill has permissions for filesystem writes (Excel and HTML report generation) and local process execution (converters).
- Sanitization: The skill does not sanitize element properties extracted from models before displaying them or writing them to files.
Audit Metadata