customer-pulse-check
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from customer reviews and CRM notes, which creates a surface for indirect prompt injection attacks.
- Ingestion points: Step 1.3 reads external files (CSV/Excel exports from Yandex Maps, 2GIS, etc.) and Step 1.2 pulls notes from CRM systems.
- Boundary markers: Absent. There are no instructions to the agent to treat this data as untrusted or to ignore embedded instructions within the content.
- Capability inventory: The skill has access to
Bash(shell execution),WebFetch(network operations), andRead(file system access). - Sanitization: Absent. There is no mention of filtering or sanitizing the imported text before processing it for theme identification.
- Mitigation: The skill includes strong "Approval Gates" requiring explicit owner confirmation before sending replies or resolving disputes, which significantly reduces the impact of a successful injection.
- [SAFE]: The skill handles sensitive business data (payment disputes, CRM entries) but follows best practices by explicitly forbidding the inclusion of personal data in summaries and requiring manual review for all transactions and communications.
Audit Metadata