The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py performs runtime code generation and process injection. It contains an embedded C source string (_SHIM_SOURCE) that is written to a temporary location and compiled using gcc to create a shared object. This object is then loaded into the environment via the LD_PRELOAD environment variable to hook system calls (socket, listen, accept) in the soffice process.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from .pptx files and extracts text content into the agent context with no sanitization or boundary protection.
Ingestion points: scripts/inventory.py extracts text from all shapes; scripts/office/unpack.py extracts raw slide XML.
Boundary markers: Absent. Extracted text is presented to the agent as plain text or structured data without "ignore embedded instructions" warnings.
Capability inventory: The skill has file-write capabilities via scripts/replace.py and arbitrary command execution via subprocess.run calls in multiple scripts.
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to execute shell commands.
scripts/office/soffice.py executes gcc to compile binary shims.
scripts/thumbnail.py and ooxml/scripts/pack.py execute soffice for document conversion.
scripts/thumbnail.py executes pdftoppm for image rendering.
[EXTERNAL_DOWNLOADS]: The skill requires multiple external dependencies from standard registries.
Python packages: markitdown[pptx], Pillow, python-pptx (implied).
Node.js packages: pptxgenjs, playwright, sharp.
System tools: soffice (LibreOffice), pdftoppm (Poppler).

pptx