render-pdf-doc

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown documents through Pandoc and XeLaTeX. While this represents a surface for indirect prompt injection (e.g., malicious instructions hidden in document text), it is an inherent property of a document processor and handled by standard academic tools.\n
  • Ingestion points: Markdown content processed in scripts/render_pdf.sh and scripts/infer_colwidths.py.\n
  • Capability inventory: File system writes (PDF), execution of Pandoc and XeLaTeX binaries, and Python-based text processing.\n
  • Boundary markers: Standard Markdown structure and Pandoc frontmatter parsing.\n
  • Sanitization: The skill relies on the parsing logic of Pandoc; it does not explicitly sanitize for LLM-specific injection patterns.\n- [DYNAMIC_EXECUTION]: The scripts/render_pdf.sh wrapper dynamically generates a temporary JSON metadata file using a Python one-liner.\n
  • Evidence: The script uses `python3
  • "$MAINFONT" "$CJKFONT"to construct the JSON, passing variables viasys.argv`.\n
  • Security Note: This implementation is secure as it avoids string interpolation into the script body, preventing shell or Python code injection.\n- [COMMAND_EXECUTION]: The skill executes external binaries (pandoc, xelatex) and shell scripts (check_deps.sh) to perform its primary rendering function. These operations are restricted to the local environment and target well-known, legitimate software.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 01:59 AM
Security Audit — agent-trust-hub — render-pdf-doc