The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script "scripts/office/soffice.py" contains an embedded C source string that is compiled using "gcc" at runtime into a shared library. This library is subsequently loaded via "LD_PRELOAD" to intercept system calls (like "socket" and "accept") in the "soffice" process to bypass environmental socket restrictions.
[PERSISTENCE]: The script "scripts/recalc.py" automatically installs a StarBasic macro ("RecalculateAndSave") into the user's LibreOffice configuration directory ("Module1.xba"). This modifies the user's persistent environment and ensures the macro is available for execution in future office sessions.
[COMMAND_EXECUTION]: The skill uses "subprocess.run" to invoke external tools including "gcc" for runtime compilation, "soffice" for spreadsheet processing, and "git" for document comparisons.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted spreadsheet data (XLSX, CSV) with significant execution capabilities.
Ingestion points: Data enters the agent context through "pd.read_excel" and "load_workbook" (referenced in "SKILL.md" and "scripts/recalc.py").
Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore instructions embedded within spreadsheet cells.
Capability inventory: The skill has access to subprocess execution, runtime code compilation, and file system writes across multiple scripts.
Sanitization: Absent; spreadsheet content is not filtered or escaped before processing.

xlsx