slides

Warn

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [DYNAMIC_EXECUTION]: The syntax highlighting helper in assets/pptxgenjs_helpers/code.js uses a dynamic require call to load PrismJS language components. The id variable, derived directly from the user-controlled lang parameter, is not validated before being used in the file path, potentially allowing for path traversal or local file inclusion (LFI) if an attacker can control the language input.
  • [COMMAND_EXECUTION]: The skill relies extensively on executing external system binaries through the subprocess and child_process modules across multiple files. While commands are generally executed using argument lists, the large variety of involved tools (including LibreOffice, Ghostscript, ImageMagick, and Inkscape) significantly increases the attack surface, particularly when processing complex, untrusted document and image formats.
  • scripts/render_slides.py and scripts/detect_font.py invoke soffice (LibreOffice) for document conversion.
  • scripts/ensure_raster_image.py invokes magick, inkscape, JxrDecApp, heif-convert, and gs (Ghostscript) for image rasterization.
  • assets/pptxgenjs_helpers/text.js invokes fc-match and fc-list for font discovery.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external content from PowerPoint (.pptx) and PDF files, which presents an attack surface for indirect prompt injection.
  • Ingestion points: Text and layout data are extracted from files in scripts/detect_font.py and rendered in scripts/render_slides.py.
  • Boundary markers: None detected; the skill processes content directly without delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill possesses broad capabilities, including file system access, network-adjacent dependencies, and extensive shell command execution.
  • Sanitization: No sanitization or filtering is performed on the extracted text before it is processed or used to influence font detection and rendering logic.
  • [METADATA_POISONING]: There is a licensing inconsistency between the skill's manifest and its legal documentation. The SKILL.md frontmatter specifies an MIT license, whereas the LICENSE.txt file contains the full text of the Apache License 2.0.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 18, 2026, 08:51 AM