omh-materials-package
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes various external document formats (PDF, XLSX, PPTX, HWP, Keynote) provided by users, creating an attack surface where malicious instructions embedded in those files could attempt to influence the agent's behavior.
- Ingestion points: User-provided source inputs and document files mentioned in the workflow (e.g., '첨부한 엑셀' (attached excel), 'PDF', 'spreadsheet', 'docx').
- Boundary markers: The instructions define a 'generation handoff boundary' and enforce an 'observed-only' policy for outcomes like binary export and render QA, which helps isolate automated processing from untrusted content.
- Capability inventory: The skill uses a suite of bundled Python scripts (pptx_read.py, pdf_read.py, xlsx_recalc.py, docx_validate.py) to extract outlines, metadata, and perform formula recalculations.
- Sanitization: No explicit content-level sanitization or prompt escaping is described for the text extracted by the reader tools, relying instead on the tools' internal parsing logic.
Audit Metadata