deal-desk
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied deal information through templates and scripts, presenting a potential vector for indirect prompt injection if the ingested data contains adversarial instructions.
- Ingestion points: External deal data is entered into assets/deal_intake_template.md and processed by scripts in the scripts/ folder.
- Boundary markers: The Python scripts generate plain text output for the agent without employing specific boundary delimiters or isolation markers to separate external data from instructions.
- Capability inventory: The included Python scripts (deal_scorer.py, discount_approval_router.py, terms_redliner.py) are limited to logical calculations and do not have access to network, file-writing, or shell execution capabilities.
- Sanitization: The scripts use standard JSON parsing and field-specific logic, which provides basic structural validation but does not filter or sanitize against natural language prompt injection content.
Audit Metadata