The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py dynamically generates a C source file in the temporary directory and executes gcc via subprocess.run to compile it into a shared object (lo_socket_shim.so).
[COMMAND_EXECUTION]: The skill uses the LD_PRELOAD environment variable to inject the dynamically compiled shared library into the soffice (LibreOffice) process to shim system socket calls.
[COMMAND_EXECUTION]: Multiple scripts (scripts/accept_changes.py, scripts/office/soffice.py) use subprocess.run to execute external binaries including soffice, pandoc, pdftoppm, and git to process documents.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection (Category 8). It ingests untrusted data from .docx files which are processed by tools like pandoc and then provided to the agent.
Ingestion points: scripts/office/unpack.py and SKILL.md (via pandoc instructions).
Boundary markers: Absent in the provided scripts; the skill relies on the agent's ability to distinguish instructions from document content.
Capability inventory: Subprocess execution of soffice, pandoc, pdftoppm, and gcc; arbitrary file read/write within the workspace.
Sanitization: The skill uses defusedxml to mitigate XML-based attacks (XXE), but does not sanitize the text content for natural language instructions.

docx