multi-agent-estimation

Pass

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external data including IFC models, PDF drawings, and project spreadsheets, which creates a surface for indirect prompt injection.
  • Ingestion points: External file inputs such as 'ifc_path', 'price_db', and drawing files referenced in the instructions and documentation.
  • Boundary markers: There are no explicit instructions or delimiters used to separate data from instructions or to warn the agent about embedded prompts.
  • Capability inventory: The orchestration logic utilizes filesystem access and network connectivity to manage construction data and interact with vector databases.
  • Sanitization: The implementation does not include specific validation or sanitization steps for the content of processed files.
  • [EXTERNAL_DOWNLOADS]: The skill documentation specifies several Python package requirements.
  • Packages: crewai, langchain-openai, ifcopenshell, pandas, and qdrant-client.
  • Status: These are reputable and well-known libraries commonly used for AI agent development and BIM data analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 5, 2026, 04:29 AM