The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess calls to execute system utilities for document processing. This includes invoking LibreOffice (soffice) for formula recalculation, git for generating text diffs during validation, and gcc for compiling a compatibility shim.
Evidence: Found in scripts/recalc.py, scripts/office/soffice.py, and scripts/office/validators/redlining.py.
[COMMAND_EXECUTION]: To support automated document saving and calculation, the skill programmatically creates a StarBasic macro file within the user's application configuration directory.
Evidence: Logic for writing Module1.xba is located in scripts/recalc.py.
[COMMAND_EXECUTION]: For operation in restricted environments where Unix sockets might be blocked, the skill generates a C-based shim at runtime and utilizes LD_PRELOAD to intercept and redirect socket calls for LibreOffice.
Implementation details are contained in scripts/office/soffice.py.
[PROMPT_INJECTION]: The skill processes untrusted external files (XLSX, DOCX, PPTX), which represents an indirect prompt injection surface. Maliciously crafted document content could potentially influence the AI agent's logic if extracted data is used in downstream reasoning tasks without strict validation.
Ingestion points include the unpacking and recalculation workflows in scripts/office/unpack.py and scripts/recalc.py.

xlsx