optimize-olap-database
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a document and table extraction pipeline (Kreuzberg Pattern) for processing unstructured operational files like PDF and DOCX reports. This ingestion of untrusted external data creates a surface for indirect prompt injection.
- Ingestion points: Technical architecture guide describes parsing tables from external PDF and DOCX documents in references/olap-database-guide.md.
- Boundary markers: There are no explicit instructions for the agent to wrap extracted content in delimiters or ignore potential instructions within the documents.
- Capability inventory: The skill is configured with broad tool access in SKILL.md, including write_file, create_file, run_build, and run_tests.
- Sanitization: While the skill mandates schema contract validation and the use of deterministic keys (SHA-256), these measures focus on data structure rather than preventing the execution of embedded natural language instructions.
Audit Metadata