The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py performs runtime compilation of C source code into a shared object file using gcc. This library is then injected into the LibreOffice (soffice) process via the LD_PRELOAD environment variable to modify system call behavior. This combination of compilation and injection represents a high-risk dynamic execution pattern used for system-level workarounds.
[COMMAND_EXECUTION]: The skill uses subprocess.run across several scripts (soffice.py, thumbnail.py, redlining.py) to execute external binaries including gcc, soffice, pdftoppm, and git for document processing and validation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PowerPoint files, creating an attack surface for indirect prompt injection via the following chain:
Ingestion points: Untrusted data enters the context through markitdown text extraction and by reading slide XML files (e.g., slide{N}.xml) during editing workflows.
Boundary markers: The instructions do not employ specific markers or delimiters to isolate extracted content from the agent's internal logic.
Capability inventory: The skill maintains the ability to execute system commands, write to the filesystem, and manipulate complex document XML.
Sanitization: Security is improved through the use of defusedxml for XML parsing and the implementation of XSD schema validation before documents are repacked.

pptx