The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py contains embedded C source code that is compiled into a shared library at runtime using gcc. The skill then injects this library into the LibreOffice (soffice) process via the LD_PRELOAD environment variable to circumvent AF_UNIX socket restrictions in sandboxed environments.
[COMMAND_EXECUTION]: Multiple Python scripts, including scripts/accept_changes.py and scripts/office/validators/redlining.py, execute system-level commands such as soffice and git using the subprocess.run() function to perform document conversions and content comparisons.
[EXTERNAL_DOWNLOADS]: The SKILL.md instructions guide the user to download and install the docx library from the public NPM registry.
[PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection because it ingests and processes untrusted Word documents using powerful system tools (soffice, pandoc, git). Instructions embedded within a processed document could potentially target the AI agent or influence the execution of the document processing pipeline.

docx