privacy-and-ai-control-applicability
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructs the agent to integrate with an external Model Context Protocol (MCP) server at 'https://account.ismscopilot.com/v1/account/mcp' to perform compliance framework lookups. While the instructions explicitly warn the agent not to include file contents, secrets, or personal data in these external queries, the use of a non-whitelisted external domain represents a potential network exfiltration vector.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted business and legal documentation to make applicability decisions.
- Ingestion points: The skill ingests data from records of processing, data processing agreements, privacy notices, vendor contracts, and AI impact assessments (SKILL.md).
- Boundary markers: The skill uses instructional constraints such as 'MUST take additional control identifiers... from the reader's licensed source' to guide the agent, though explicit delimiters for the external content are not defined in the prompt template.
- Capability inventory: The skill has the capability to write to the local file system (creating './privacy-and-ai-control-applicability/') and perform network lookups via the MCP server (SKILL.md).
- Sanitization: The instructions include explicit prohibitions ('NEVER copy personal data', 'Send only a short question in your own words') to prevent the direct relay of untrusted content into sensitive outputs or external requests.
Audit Metadata