The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains an embedded C source string that is written to a temporary location and compiled at runtime into a shared library using gcc.
[PRIVILEGE_ESCALATION]: The skill uses the LD_PRELOAD environment variable to inject the compiled shim library into the soffice execution environment. While intended for sandbox compatibility (shimming AF_UNIX sockets), process injection is a high-risk technique.
[DYNAMIC_EXECUTION]: In scripts/accept_changes.py, the skill dynamically generates a LibreOffice Basic macro (Module1.xba) and saves it to a temporary profile directory to automate document tasks.
[COMMAND_EXECUTION]: The skill relies on several external CLI tools including soffice, pandoc, gcc, and pdftoppm for document conversion, validation, and shim compilation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user documents which serves as an attack surface. 1. Ingestion points: Reads XML content from user-provided .docx files. 2. Boundary markers: None identified in the prompt instructions to delineate or isolate untrusted document content. 3. Capability inventory: Spawns shell commands, compiles binary code, and modifies the local filesystem. 4. Sanitization: Employs defusedxml to protect against XML-based attacks during parsing.

docx