The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains an embedded C source string that is written to the temporary directory and compiled at runtime using gcc. The resulting shared object is then used via LD_PRELOAD to shim system calls (socket, listen, accept) to allow LibreOffice to function in restricted networking environments. Additionally, scripts/recalc.py dynamically writes a StarBasic macro to the LibreOffice profile to automate the recalculation process.
[COMMAND_EXECUTION]: The skill performs several subprocess calls using subprocess.run. It executes soffice for spreadsheet recalculation, gcc for compiling its internal shim, and git for generating word-level diffs in the tracked-changes validator.
[EXTERNAL_DOWNLOADS]: The skill's instructions in SKILL.md direct the agent to use pip install to download and install standard libraries (openpyxl, pandas, markitdown) if they are missing from the environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes and recalculates data from untrusted spreadsheets, creating a potential vector for indirect prompt injection through malicious formulas or content in user-provided files.
Ingestion points: scripts/recalc.py and the Office XML validators in scripts/office/validators/ which parse external document components.
Boundary markers: Absent; there are no instructions to use specific delimiters when interpolating data into models.
Capability inventory: The skill has the capability to execute system binaries, compile C code, and rewrite spreadsheet files on disk.
Sanitization: The skill mitigates common data-processing risks by using defusedxml to prevent XXE attacks and implementing strict path-traversal checks in its archive extraction logic (safe_extract in scripts/office/helpers/__init__.py).

xlsx