The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py contains a dynamic execution pattern where it compiles C code at runtime. It writes a hardcoded C source string (_SHIM_SOURCE) to a local file in the user's cache directory, then invokes the gcc compiler to produce a shared object library (lo_socket_shim.so). It utilizes the LD_PRELOAD environment variable to inject this shared library into the LibreOffice (soffice) process to shim socket-related system calls. Although this is designed to facilitate formula recalculation in restricted environments, runtime compilation and process injection are high-risk behaviors.
[COMMAND_EXECUTION]: Several components of the skill execute system-level commands through Python's subprocess module. scripts/recalc.py and scripts/office/soffice.py execute soffice for recalculating workbook formulas. scripts/office/soffice.py executes the gcc compiler. scripts/office/validators/redlining.py executes git to perform document diffing.
[PROMPT_INJECTION]: The skill processes untrusted Excel workbooks and CSV/TSV data provided by users or retrieved from the environment. It has access to powerful tools including file system operations (Read, Write, Edit) and shell access (Bash). The combination of data ingestion and high-privilege tool access creates a vulnerability surface for indirect prompt injection, where malicious instructions hidden in a spreadsheet could potentially trigger unintended actions by the agent.

xlsx