skills/agentscope-ai/qwenpaw/pdf/Gen Agent Trust Hub

pdf

Pass

Audited by Gen Agent Trust Hub on Apr 28, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill processes external PDF files and extracts text content using libraries like pypdf and pdfplumber. This text is then provided to the agent. If a PDF contains hidden or adversarial instructions, the agent might execute them as if they were valid directives.
  • Ingestion points: Text and table extraction logic in scripts/extract_form_structure.py, scripts/extract_form_field_info.py, and the main SKILL.md guide.
  • Boundary markers: No specific delimiters or instructions to ignore embedded commands are implemented in the provided scripts or prompts.
  • Capability inventory: The skill utilizes file write operations (pypdf.PdfWriter) and encourages the use of shell command execution for its scripts.
  • Sanitization: There is no evidence of text validation or sanitization to prevent malicious content from influencing the agent's behavior.
  • [COMMAND_EXECUTION]: The skill relies on local script execution and command-line utilities (qpdf, poppler-utils, ImageMagick). While these are standard tools, they are invoked with arguments (filenames and paths) that could be manipulated if not properly handled by the agent's shell execution environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 28, 2026, 05:35 AM