The Agent Skills Directory

[COMMAND_EXECUTION]: The file scripts/office/soffice.py dynamically generates and compiles a C source file using gcc. The resulting shared library is injected into subsequent processes via the LD_PRELOAD environment variable to shim socket functions. This mechanism is intended to facilitate communication for LibreOffice in sandboxed environments.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection (Category 8) by processing untrusted external data with high execution capabilities.
Ingestion points: Untrusted data enters the context via the extraction and visual analysis of user-provided PPTX files in scripts/office/unpack.py, scripts/thumbnail.py, and scripts/office/pack.py.
Boundary markers: Absent. The scripts do not implement delimiters or explicit instructions to ignore embedded commands in extracted content.
Capability inventory: High. Includes the ability to execute shell commands, compile C code (gcc), and run external binaries (soffice, pdftoppm).
Sanitization: The skill correctly utilizes defusedxml for XML parsing, which mitigates XML-based attacks but does not sanitize the natural language content for instruction injection.
[EXTERNAL_DOWNLOADS]: Documentation in SKILL.md and pptxgenjs.md requires the installation of several external Python and Node.js packages, including markitdown, Pillow, defusedxml, lxml, pptxgenjs, react, and sharp.

pptx