The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data in a wide variety of formats (e.g., PDF, Office documents, HTML, LaTeX, and Jupyter Notebooks) via the visualize and read_video tools. This represents a significant attack surface where maliciously crafted files could contain instructions designed to influence or override agent behavior.
Ingestion points: Local files processed by the visualize, read_image, and read_video tools defined in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or safety prompts advising the agent to treat data returned by these tools as untrusted.
Capability inventory: The tools provide extensive file reading, metadata extraction, rendering, and manipulation capabilities.
Sanitization: The provided documentation does not specify the use of sanitization or validation logic for the content extracted from external files.
[COMMAND_EXECUTION]: The visualize tool depends on external system utilities to process specific file types, such as libreoffice for Office documents, playwright for web page screenshots, and blender for 3D models. Additionally, LaTeX support involves the compilation of .tex files, which traditionally requires invoking system-level typesetting engines.
[EXTERNAL_DOWNLOADS]: The skill documentation identifies several external software requirements and dependencies, including libreoffice, playwright, blender, pypdfium2, and resvg, which are necessary for the visualization of various document and media formats.

qwen-mm-plugins-core