pdf-to-image-preview
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The
references/usage-guide.mdfile contains instructions that are inconsistent with the provided implementation. - Evidence: The guide refers to a script named
pdf_to_images.pyand a parameter--html-output, whereas the actual file isscripts/convert_pdf_to_images.pyand does not support HTML output. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (PDF files) which presents a theoretical attack surface for indirect prompt injection.
- Ingestion points: The script reads PDF files provided by the user or external sources via
fitz.open(pdf_path)inscripts/convert_pdf_to_images.py. - Boundary markers: None present in the data processing flow.
- Capability inventory: The skill can write files to the local filesystem and create directories via
os.makedirsandpix.save. - Sanitization: No explicit sanitization or content filtering is performed on the PDF structure; however, the script only performs image rendering and does not return extracted text to the agent, significantly mitigating the risk of instructions influencing agent logic.
Audit Metadata