skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/change-order-processor/Gen Agent Trust Hub
change-order-processor
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided CSV, Excel, and JSON data as described in instructions.md. This represents a potential indirect prompt injection surface as user data is processed by the agent. Although the skill has file system write access via the export_log method in SKILL.md, the lack of network access limits exfiltration risk. Evidence chain: 1. Ingestion points: instructions.md (CSV, Excel, JSON). 2. Boundary markers: Absent. 3. Capability inventory: File system write (SKILL.md, export_log). 4. Sanitization: Absent.
- [SAFE]: Technical implementation in SKILL.md uses standard Python libraries (pandas, openpyxl) for data processing and report generation without any malicious behavior.
Audit Metadata