bel-crm-db
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external, untrusted sources such as PostgreSQL database records and MS365 emails which may contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: Data is ingested from text fields like
person.notes,sales_opportunity.description,event.description,data_files.email_body_text, anddata_files.extracted_text(SKILL.md). - Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat retrieved data as untrusted or to ignore embedded commands.
- Capability inventory: The agent is granted
mcp__postgresql__write_queryandmcp__ms365access, allowing it to modify data or send emails based on potentially poisoned input. - Sanitization: The instructions lack requirements for sanitizing, escaping, or validating the content of emails or database fields before processing them.
- [DATA_EXPOSURE]: The skill defines a schema for storing high-value sensitive data, creating a significant target for exfiltration if the agent is compromised.
- PII Exposure: The
adressen,person, andcompany_sitetables store comprehensive personal and corporate contact details, including names, private emails, phone numbers, and physical addresses. - Sensitive File Access: The
data_filestable explicitly stores full file contents (PDFs, Office documents, email bodies) in a Base64-encodedfile_datafield, which may contain trade secrets or confidential communications.
Audit Metadata