The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs runtime compilation and process injection to enable LibreOffice functionality in restricted environments.
The file scripts/office/soffice.py contains embedded C source code that implements a socket shim.
At runtime, the skill uses gcc to compile this C code into a shared object (.so) file stored in the temporary directory.
It then uses the LD_PRELOAD environment variable to inject this shim into the soffice process during execution.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute several system-level command-line tools.
It calls soffice to process document changes and conversions.
It calls gcc to compile its internal compatibility shim.
It calls git diff to perform text-based comparisons during document validation.
[INDIRECT_PROMPT_INJECTION]: The skill processes complex external data structures which presents an indirect injection surface.
Ingestion points: Office documents (.docx, .pptx, .xlsx) are unpacked, and their XML and relationship files are read into the agent context (in scripts/office/unpack.py).
Boundary markers: Absent. The instructions do not explicitly tell the agent to treat document content as untrusted text or to ignore embedded instructions.
Capability inventory: The skill possesses significant capabilities including arbitrary file writing, execution of CLI tools, and runtime code compilation.
Sanitization: The skill mitigates common XML attacks by using the defusedxml library for parsing document components.

docx