The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains a C source code string (_SHIM_SOURCE) which it writes to disk, compiles at runtime using gcc, and then injects into the process environment using LD_PRELOAD. This is designed to shim AF_UNIX socket calls to support LibreOffice in sandboxed or restricted execution environments.
[DYNAMIC_EXECUTION]: The script scripts/accept_changes.py dynamically generates a LibreOffice Basic macro and writes it to a temporary user profile (/tmp/libreoffice_docx_profile) to automate the acceptance of tracked changes in documents.
[COMMAND_EXECUTION]: Multiple scripts (accept_changes.py, soffice.py, redlining.py) utilize subprocess.run to execute external system commands including soffice (LibreOffice), gcc, git, and pdftoppm.
[INDIRECT_PROMPT_INJECTION]: The skill accepts untrusted .docx and .dotx files as input (ingestion points in scripts/merge_runs.py and scripts/office/validate.py). While the skill employs security best practices such as using defusedxml and explicitly instructing the agent to delete symlinks in unpacked directories, the complexity of the Office Open XML schema combined with high-privilege capabilities creates an attack surface for indirect injection.
[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md direct the agent to utilize external dependencies including the docx npm package and Python libraries like lxml and defusedxml if not already present.

docx