sql-optimization-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied SQL source material and external data, creating an attack surface for indirect prompt injection.
- Ingestion points: User-provided stack/source material and external datasets fetched via sandbase_run (SKILL.md).
- Boundary markers: Instructions explicitly state to 'Preserve the user's chosen stack, source material, and authorization boundaries' and 'Send only the minimum necessary data.'
- Capability inventory: The skill instructions cover SQL implementation, analysis, and execution plan inspection (SKILL.md).
- Sanitization: Includes quality gates to inspect data shapes, types, and validate outcomes against independent baselines.
- [EXTERNAL_DOWNLOADS]: The skill uses vendor-specific tools to access external datasets and model inference capabilities.
- Evidence: Tools such as sandbase_run and sandbase_inspect are used to interact with external services (references/sandbase-api-map.md). The skill mandates a verification step where the agent must present the endpoint, inputs, and costs to the user for approval before execution.
Audit Metadata