omh-deliverable-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 document formats including PPT, PDF, XLSX, DOCX, HWP, and Markdown, creating a surface for instructions embedded in these files to influence agent behavior.
- Ingestion points: Reading document structures and content using tools such as
pptx_read.py,pdf_read.py,xlsx_read.py, anddocx_read.py(SKILL.md). - Boundary markers: The skill documentation explicitly defines a 'prepared-vs-observed boundary' and instructs the agent to separate guidance from observed evidence, which acts as a mitigation.
- Capability inventory: The skill uses tools to render screenshots, calculate formulas, and validate package health, which could be abused if content is maliciously crafted.
- Sanitization: No specific sanitization or filtering of text content within the ingested documents is described, relying instead on manual inspection of rendered images.
Audit Metadata