The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation and process injection during its operation. The script scripts/office/soffice.py writes C source code to a temporary file and compiles it into a shared library using the system's gcc compiler. This library is then injected into the LibreOffice (soffice) process via the LD_PRELOAD environment variable to facilitate socket management in restricted environments. Additionally, scripts/recalc.py creates a LibreOffice macro file (Module1.xba) at runtime to automate model recalculations.
[COMMAND_EXECUTION]: The skill utilizes the subprocess module to execute several external system binaries required for spreadsheet processing: soffice is invoked for formula evaluation, gcc is used to compile the compatibility shim, and git is used by the redlining validator to compare document versions. These commands are executed with structured argument lists to prevent shell injection.
[PROMPT_INJECTION]: The skill ingests untrusted spreadsheet data via the pandas and openpyxl libraries (referenced in scripts/recalc.py and SKILL.md) without utilizing explicit boundary markers to isolate data from instructions. The skill possesses capabilities for command execution and file writing, creating a surface for potential indirect prompt injection. While the use of defusedxml provides protection against XML-level exploits like XXE, the skill does not include specific logic to sanitize or ignore natural language instructions embedded within spreadsheet cells.

xlsx