deal-desk
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface by ingesting external data (YAML deal specs and CSV CRM exports) which are processed by included Python scripts to generate reports. This is a common pattern for operational tools.
- Ingestion points: Untrusted data enters the agent context via
scripts/deal_review_packet.py(YAML),scripts/deal_velocity_analyzer.py(CSV), andscripts/discount_authority_router.py(YAML). - Boundary markers: None explicitly defined in the scripts, though outputs are clearly labeled as reports.
- Capability inventory: No network operations, sensitive file access, or subprocess execution found in any script.
- Sanitization: The scripts use a custom-built, restrictive YAML parser and the standard Python CSV library, preventing common deserialization attacks.
- [COMMAND_EXECUTION]: The skill includes Python scripts intended for local execution to automate report generation. These scripts are well-structured, use only standard library modules, and do not perform any privileged or dangerous system operations.
Audit Metadata