The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py performs runtime compilation and library injection. It writes an embedded C source string to a temporary file, compiles it using gcc, and loads the resulting shared object via the LD_PRELOAD environment variable to bridge socket calls for LibreOffice.
[COMMAND_EXECUTION]: Multiple scripts execute system utilities using subprocess.run, including gcc for the socket shim, soffice for PDF conversion, pdftoppm for image extraction, and git for calculating document differences. These tools are powerful and used to support the skill's primary document processing functions.
[EXTERNAL_DOWNLOADS]: The skill requires several third-party dependencies from standard package registries. These include Python packages like markitdown, Pillow, defusedxml, and lxml, as well as Node.js packages such as pptxgenjs, react-icons, and sharp.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests and processes untrusted data from external .pptx files.
Ingestion points: XML content is read from user-provided files in scripts/office/unpack.py and scripts/thumbnail.py.
Boundary markers: Absent; the skill does not use specific delimiters or instructions to prevent the agent from following commands embedded in the slide XML.
Capability inventory: High; the skill can execute system commands (gcc, soffice, git) and perform file system operations.
Sanitization: Partial; while the skill uses defusedxml to mitigate XML-specific attacks like XXE, it does not sanitize plain text content against injection.

pptx