zg-private-chat-sales-assistant
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns or security risks were identified in the skill's instructions or referenced documentation.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, including customer chat histories and third-party knowledge bases provided as attachments. This is categorized as safe because the skill lacks exploitable capabilities such as network access, file system write permissions, or shell command execution. Additionally, the instructions provide clear boundary markers for the agent to distinguish between customer input, facts, and assumptions.
- Ingestion points: Processes chat history and knowledge base files via the agent's file-reading capabilities (SKILL.md).
- Boundary markers: Explicit instructions to separate 'Customer's exact words', 'Known facts', and 'Potential concerns' (SKILL.md).
- Capability inventory: None detected; the skill only generates text responses and does not invoke external tools for operations beyond text analysis.
- Sanitization: No explicit sanitization logic is provided, relying on the underlying model's safety filters.
Audit Metadata