documentdb-natural-language-querying
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill establishes extensive safety guidelines for the handling of sensitive database documents. It defines rigorous redaction rules for field names (e.g., password, apikey, secret) and data patterns (e.g., JWTs, connection strings, PEM blocks) to prevent accidental data exposure.
- [SAFE]: No malicious patterns such as prompt injection, obfuscation, or unauthorized credential harvesting were identified. The instructions prioritize security by enforcing schema-only inference and prohibiting the use of sampled data in generated output.
- [SAFE]: The skill utilizes the Model Context Protocol (MCP) framework and restricts its tool usage to specifically scoped database operations (
mcp__documentdb__*), following the principle of least privilege. - [SAFE]: External references are directed to official Microsoft documentation for Azure DocumentDB, which is a verified and reputable source.
- [SAFE]: Indirect Prompt Injection Assessment: The skill has an ingestion surface for untrusted data (database documents accessed via MCP tools in SKILL.md). While explicit boundary markers in the data stream are not defined, the skill mandates strict sanitization and redaction of this data. Additionally, the skill's capabilities are limited to generating read-only query structures with no access to shell execution or external network operations, effectively mitigating potential injection risks.
Audit Metadata