skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/co2-carbon-footprint/Gen Agent Trust Hub
co2-carbon-footprint
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (BIM model quantities) which creates a surface for indirect prompt injection.
- Ingestion points: Data is ingested through file paths (Excel/CSV) and interpolated into the processing logic via pandas DataFrames as described in
SKILL.mdandinstructions.md. - Boundary markers: The instructions lack explicit boundary markers or delimiters to separate the data from the agent's internal instructions, which could allow instructions embedded in metadata fields (like material names) to influence agent behavior.
- Capability inventory: The skill possesses filesystem write capabilities (
export_resultsto Excel) and complex data processing logic usingpandas. - Sanitization: The Python implementation includes type-casting for quantitative values (e.g.,
float(element.get('volume'))), which provides basic protection against non-numeric payloads in calculation fields.
Audit Metadata