kql-grounded-queries
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- External Documentation Grounding: The skill utilizes official documentation pages to verify the existence of tables and columns before generating queries. This grounding process involves fetching content from external learn.microsoft.com URLs to ensure identifiers are accurate and not hallucinated.
- Third-Party Service Integration: The skill provides instructions for using an optional community-hosted search service at kqlsearch.com. The documentation clearly identifies this as an external, non-vendor service that is configured by the user and managed under its own terms.
- Data Privacy Safeguards: The instructions include robust constraints to prevent sensitive information—such as account names, IP addresses, and hostnames—from being included in external search strings. This ensures that only public table names and generic keywords are sent to services outside the tenant environment.
- Indirect Prompt Injection Surface: By ingesting community-provided KQL snippets from external sources, the skill introduces a potential surface for indirect prompt injection. However, it explicitly mitigates this risk with instructions to treat all retrieved content as data rather than instructions, ensuring that embedded directives are ignored by the agent.
Audit Metadata