The Agent Skills Directory

[COMMAND_EXECUTION]: Several scripts (pptx_render.py, xlsx_recalc.py) and test files (test_docx_skill.py, test_powerpoint_skill.py, test_xlsx_skill.py) utilize subprocess.run to invoke external system tools such as soffice (LibreOffice), pdftoppm, and pdftocairo. These operations are conducted using list-based arguments for safety and are integral to the skill's primary functions, such as rendering slides to images and recalculating Excel formulas.
[EXTERNAL_DOWNLOADS]: The SKILL.md prerequisites section contains a shell script that installs specific, pinned versions of well-known Python libraries (e.g., python-docx, openpyxl, python-pptx) from the standard PyPI registry. This is a standard and expected procedure for managing the skill's dependencies.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to extract content from user-provided Office documents, it inherently possesses an attack surface for indirect prompt injection. Malicious instructions embedded in a processed document could be interpreted by the agent during the reading phase.
Ingestion points: docx_read.py, xlsx_read.py, and pptx_read.py are used to extract text and structure from documents.
Boundary markers: No explicit markers are used to delimit extracted content from instructions.
Capability inventory: The skill can perform file system writes and execute subprocesses through various utility scripts.
Sanitization: The scripts do not perform specific sanitization or filtering of the extracted text content.

microsoft-office