job-sales
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements strong security boundaries (Red Lines) that prohibit autonomous actions such as sending messages, submitting quotes, or granting discounts without human intervention.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including customer chat histories and meeting recordings, creating a potential surface for indirect prompt injection.
- Ingestion points: Customer chat logs and meeting recordings (SKILL.md).
- Boundary markers: The skill uses negative constraints (e.g., "do not speculate", "write 'no record' if missing") to prevent the model from acting on missing or potentially manipulated data.
- Capability inventory: Accesses external business connectors (Tianyancha, QixinHuiyan, ShuidiXinyong) and performs writes to multidimensional tables (Bitable) for CRM updates.
- Sanitization: Requires mandatory human review for all CRM field updates and forbids autonomous external communication.
Audit Metadata