churn-lost-deal-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from a CRM, which creates a surface for indirect prompt injection.
- Ingestion points: The skill reads deal notes, call notes, and comments from a CRM (e.g., HubSpot, Attio) via an MCP tool.
- Boundary markers: The prompt template lacks explicit delimiters or instructions for the agent to ignore potentially malicious content embedded within the CRM notes.
- Capability inventory: The skill has the ability to read from the CRM and write reports to Google Drive.
- Sanitization: No evidence of sanitization or filtering of the ingested CRM content is provided in the instructions.
- Recommendation: Implement boundary markers (e.g., XML tags or clear separators) when interpolating CRM data and instruct the agent to treat the content as data only, ignoring any embedded commands.
Audit Metadata