The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several system commands through the Python subprocess module and performs runtime compilation.
It uses gcc to compile an embedded C source string into a shared library (lo_socket_shim.so) at runtime.
It performs process injection by setting the LD_PRELOAD environment variable to load this custom library into the soffice (LibreOffice) process space to modify its networking behavior.
It executes system utilities including soffice for document conversion, pdftoppm for image generation, and git for generating document diffs.
[PROMPT_INJECTION]: The skill processes untrusted PowerPoint files, which creates a surface for indirect prompt injection attacks.
Ingestion points: scripts/office/unpack.py extracts and parses XML content from user-provided .pptx and .docx files.
Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands within the ingested text.
Capability inventory: The skill possesses high-risk capabilities, including arbitrary command execution and file system access.
Sanitization: While defusedxml is used to prevent XML-based attacks, there is no validation or filtering of natural language instructions that might be embedded in slide content.
[EXTERNAL_DOWNLOADS]: The skill documentation references multiple external dependencies required for full functionality.
Python dependencies: markitdown[pptx], Pillow, defusedxml, and lxml.
Node.js dependencies: pptxgenjs, react-icons, react, react-dom, and sharp.
System requirements: LibreOffice and Poppler utility suite.

pptx