The Agent Skills Directory

[DYNAMIC_EXECUTION]: The module scripts/office/soffice.py contains a hardcoded C source string (_SHIM_SOURCE) that is written to a temporary file and compiled at runtime using gcc into a shared library.
[PRIVILEGE_ESCALATION]: The script scripts/office/soffice.py load the compiled C shared library into the soffice (LibreOffice) process via the LD_PRELOAD environment variable. This technique intercepts system-level socket calls and represents a form of process injection that could be abused to execute arbitrary code with the permissions of the office suite.
[DYNAMIC_EXECUTION]: The script scripts/accept_changes.py dynamically constructs a LibreOffice StarBasic macro and writes it to a temporary user profile (/tmp/libreoffice_docx_profile) to automate tracked changes acceptance.
[COMMAND_EXECUTION]: Multiple scripts (scripts/formula.py, scripts/accept_changes.py, scripts/office/soffice.py) use subprocess.run to execute external binary tools including pandoc, gcc, and soffice with arguments derived from runtime processing.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: scripts/office/unpack.py and scripts/new_doc.js ingest untrusted user content from .docx, .doc, and .md files.
Boundary markers: Absent; file content is parsed and processed into XML structures without isolation from the agent's control context.
Capability inventory: The skill possesses high-privilege capabilities including binary compilation, library injection, and macro execution across multiple files.
Sanitization: Basic XML entity escaping is present, but there is no structural sanitization or validation of embedded logic within the processed documents.

docx-editor-cn