The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes shell commands through Python's subprocess.run module to perform its tasks.
scripts/accept_changes.py executes soffice to process documents.
scripts/office/soffice.py executes gcc to compile a shared library from source code at runtime.
scripts/office/validators/redlining.py executes git diff to compare text content.
[DYNAMIC_EXECUTION]: scripts/office/soffice.py generates and executes code dynamically at runtime.
It contains a C source code string (_SHIM_SOURCE) which is written to a temporary file (lo_socket_shim.c).
It compiles this source into a shared object file (lo_socket_shim.so) using gcc at runtime.
It uses the LD_PRELOAD environment variable to inject this compiled shim into the soffice process, hooking system calls like socket, listen, and accept.
[EXTERNAL_DOWNLOADS]: SKILL.md provides instructions for the user to install external dependencies required for the skill's operation.
Recommends installing the docx JavaScript library globally using npm install -g docx.
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection (Category 8).
Ingestion points: The skill reads and processes untrusted .docx files provided by users via scripts/office/unpack.py.
Boundary markers: There are no explicit markers or instructions to ignore potential commands embedded within the document XML.
Capability inventory: The skill possesses extensive capabilities including file system access, arbitrary shell command execution, and runtime code compilation.
Sanitization: While the skill uses defusedxml to prevent XML-based attacks (XXE), the extracted text is processed by an LLM which may follow instructions embedded in the document text.

docx