The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py generates C source code at runtime and compiles it into a shared object using gcc. The resulting library is injected into the LibreOffice environment using LD_PRELOAD to shim socket functions. This technique of runtime compilation and process injection is a high-risk pattern used to bypass environment restrictions.
[COMMAND_EXECUTION]: Multiple scripts invoke external system commands via the subprocess module. Notable instances include scripts/office/soffice.py (executing gcc and soffice), scripts/accept_changes.py (executing soffice), and scripts/office/validators/redlining.py (executing git diff). These commands are used for document processing and validation.
[EXTERNAL_DOWNLOADS]: The markdown body of SKILL.md requires the installation of the docx library from the npm registry.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by unzipping and parsing user-supplied .docx files.
Ingestion points: scripts/office/unpack.py extracts XML content from document archives.
Boundary markers: Absent; the skill does not use delimiters to isolate document content from instructions.
Capability inventory: The skill possesses broad capabilities, including arbitrary command execution via soffice and gcc.
Sanitization: The skill employs defusedxml for XML parsing, providing mitigation against XML external entity (XXE) vulnerabilities.

docx