skills/espennilsen/pi/pdf-reader/Gen Agent Trust Hub

pdf-reader

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes content from external PDF files which may contain instructions targeting the AI agent.\n
  • Ingestion points: Extracted text is read by the agent from pdftotext output and the scripts/extract.py script.\n
  • Boundary markers: Output contains minimal separation (e.g., '--- Page 1 ---'), which does not provide a robust security boundary to prevent instruction obedience.\n
  • Capability inventory: The agent is tasked with summarizing and analyzing the extracted text, creating a direct path for injection.\n
  • Sanitization: No sanitization or filtering is performed on the text extracted from the PDF.\n- [EXTERNAL_DOWNLOADS]: The skill downloads PDF files from user-provided URLs.\n
  • Evidence: curl -sL "URL" -o /tmp/document.pdf in SKILL.md.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute various command-line tools for document processing.\n
  • Evidence: Uses pdftotext, pdfinfo, pdfimages, and python3 scripts/extract.py.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 03:41 PM