The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py performs dynamic code generation and compilation at runtime.
Evidence: The script writes C source code to a temporary file and executes gcc -shared -fPIC -o ... to compile it.
Context: This is intended to shim socket operations for LibreOffice in restricted environments, but runtime compilation is a high-risk behavior.
[COMMAND_EXECUTION]: The script scripts/office/soffice.py uses library injection to modify process behavior.
Evidence: The script sets the LD_PRELOAD environment variable to the path of the compiled shared object before running soffice.
Context: This is used to hook system calls in the LibreOffice process to bypass environment limitations.
[EXTERNAL_DOWNLOADS]: The skill instructions require the user to install external global dependencies.
Evidence: npm install -g docx in SKILL.md.
[COMMAND_EXECUTION]: Multiple scripts execute shell commands and external binaries with potentially untrusted arguments.
Evidence: scripts/accept_changes.py and scripts/office/validators/redlining.py use subprocess.run to call soffice and git.
Context: While arguments are partially sanitized, the execution of complex external binaries increases the attack surface.

docx