The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied SVG files, which serves as an entry point for untrusted data that could contain malicious instructions or payloads.
Ingestion points: SVG files are read from the filesystem in scripts/convert.py using source.read_bytes().
Boundary markers: The skill does not implement natural language boundary markers or "ignore" instructions for the data it processes.
Capability inventory: The skill invokes several external binaries (Chrome, LibreOffice, MuPDF, Poppler) via subprocess.Popen and performs file system operations in scripts/convert.py, scripts/native.py, and scripts/check.py.
Sanitization: scripts/convert.py implements a check_self_contained function that explicitly rejects SVGs containing <script> tags, <foreignObject> tags, event handlers (e.g., onclick), and external resource references (e.g., href, src, or CSS @import). Additionally, the xml function in the same file uses lxml.etree.XMLParser with resolve_entities=False and no_network=True to prevent XXE attacks.
[COMMAND_EXECUTION]: The conversion pipeline relies on the execution of multiple system utilities.
Evidence: scripts/convert.py executes google-chrome, mutool, soffice, pdfimages, and pdftoppm using subprocess.Popen. These calls are constructed as argument lists rather than shell strings, which reduces the risk of command injection.
[EXTERNAL_DOWNLOADS]: The setup script installs necessary Python dependencies from official package registries.
Evidence: scripts/install.sh uses pip to install the python-pptx, Pillow, fonttools, and lxml packages.

editable-vector-slides