The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Runtime generation and injection of executable code.
The script scripts/office/soffice.py contains raw C source code in the _SHIM_SOURCE variable.
At runtime, this code is written to a temporary file (/tmp/lo_socket_shim.c) and compiled into a shared library using gcc.
The skill then uses the LD_PRELOAD environment variable to inject this compiled library into the soffice process to shim network-related system calls.
[COMMAND_EXECUTION]: Use of the subprocess module to execute system binaries.
scripts/recalc.py executes the soffice binary with macro parameters to trigger formula recalculation.
scripts/office/soffice.py executes gcc to compile internal shims at runtime.
scripts/office/validators/redlining.py invokes git diff to perform text comparisons between document versions.
[PROMPT_INJECTION]: Vulnerability surface for indirect prompt injection through ingestion of untrusted files.
Ingestion points: Untrusted data from spreadsheet files is loaded into the agent context via pandas.read_excel (as shown in SKILL.md examples) and openpyxl.load_workbook (in scripts/recalc.py).
Boundary markers: None identified; spreadsheet data is processed without delimiters or markers to distinguish untrusted content from system instructions.
Capability inventory: The skill possesses shell execution capabilities through multiple scripts including scripts/recalc.py and scripts/office/soffice.py.
Sanitization: No sanitization or safety checks are performed on the content of processed spreadsheets before they influence agent behavior.

xlsx