The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py executes gcc via subprocess.run to compile a custom C shim into a shared library. This compilation occurs at runtime if the environment detects restricted socket access.
[COMMAND_EXECUTION]: Multiple scripts, including recalc.py and scripts/recalc.py, use subprocess.run to call the soffice (LibreOffice) binary to perform formula recalculations in spreadsheet files.
[COMMAND_EXECUTION]: The validation logic in scripts/office/validators/redlining.py executes the git command via subprocess.run to generate word-level diffs for document version comparison.
[REMOTE_CODE_EXECUTION]: In scripts/office/soffice.py, the skill performs dynamic execution by compiling C source code and injecting the resulting shared object into the LibreOffice process using the LD_PRELOAD environment variable. This technique is used to bypass sandbox restrictions on Unix sockets.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted Office documents (Excel, Word, PowerPoint) and provides their content to the AI agent context. Maliciously crafted documents could contain hidden instructions designed to bypass agent constraints or influence its behavior during data processing.

xlsx