The Agent Skills Directory

[DYNAMIC_EXECUTION]: The module scripts/office/soffice.py contains embedded C source code in the _SHIM_SOURCE variable. The _ensure_shim function writes this source to a temporary location and executes gcc -shared -fPIC to compile it into a shared object. The get_soffice_env function then utilizes LD_PRELOAD to inject this library into the soffice process at runtime to override standard socket functions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Excel workbooks and provides a significant capability surface, establishing an attack vector for indirect prompt injection.
Ingestion points: untrusted Excel files are loaded via openpyxl and pandas in SKILL.md, and validated via lxml in scripts/office/validators/base.py.
Boundary markers: Absent. There are no instructions to use specific delimiters or ignore embedded instructions when reading cell values or sheet names.
Capability inventory: The skill executes several subprocesses including soffice, gcc, git, and markitdown across scripts/recalc.py, scripts/office/soffice.py, and scripts/office/validators/redlining.py.
Sanitization: Absent. Data from spreadsheets is used to construct paths and passed into processing logic without visible filtering.
[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to run system commands. Specifically, scripts/recalc.py and scripts/office/soffice.py launch soffice with complex arguments, and scripts/office/validators/redlining.py invokes git diff on file contents.

xlsx