skills/nousresearch/hermes-agent/pdf/Gen Agent Trust Hub

pdf

Pass

Audited by Gen Agent Trust Hub on Aug 30, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the pdftoppm utility via subprocess.run to rasterize PDF pages. The implementation is secure as it passes arguments as a list without invoking a shell, which effectively prevents shell injection vulnerabilities.
  • [DYNAMIC_EXECUTION]: The scripts utilize __import__ for lazy loading of optional library dependencies such as pdfplumber and pypdf. These calls are restricted to hardcoded, static module names and do not allow for arbitrary dynamic code execution.
  • [DATA_EXFILTRATION]: The metadata management tool includes functionality to extract embedded PDF attachments. It mitigates potential path traversal (ZipSlip) attacks by applying os.path.basename to all attachment filenames before writing them to the host file system.
  • [INDIRECT_PROMPT_INJECTION]: While the skill's primary function involves ingesting external document data, it manages this risk by providing structured JSON output for all extraction tasks and specific instructions for the agent to handle unformatted or image-based content through designated fallback skills.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 30, 2026, 02:23 PM
Security Audit — agent-trust-hub — pdf