find-customer-quotes
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFE
Full Analysis
- Data Source Interaction: The skill integrates with internal sources like meeting notes and CRM to retrieve context. This interaction is managed through a preflight configuration step, ensuring structured and authorized data access.
- External Data Ingestion: The skill processes potentially untrusted transcript data. To mitigate risks associated with indirect prompt injection, it uses a two-pass extraction and verification process that focuses on verbatim snippets and assesses speaker role likelihood using grounded evidence.
- Safety Guardrails for Outgoing Actions: The skill includes explicit instructions to seek user approval before performing any permanent or external actions such as saving, posting, or sharing quote packages. This ensures human oversight of all generated outputs.
- Verification Heuristics: It employs detailed heuristics to identify speaker roles (e.g., customer vs. internal seller), which helps prevent the misattribution or fabrication of customer statements.
- Modular Logic: The use of external reference files for detailed extraction and output rules facilitates clear instruction boundaries and consistent logic application.
Audit Metadata