The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py contains a hardcoded C source code string that is written to a temporary file and compiled into a shared library using gcc at runtime. This shared library is then loaded using LD_PRELOAD to shim socket calls during LibreOffice execution. Additionally, scripts/accept_changes.py dynamically generates a LibreOffice Basic macro and saves it to a temporary user profile at runtime to facilitate the automated acceptance of tracked changes.
[COMMAND_EXECUTION]: Multiple Python scripts perform system command execution using subprocess.run. For example, scripts/office/soffice.py executes gcc for library compilation and soffice for document processing, while scripts/office/validators/redlining.py invokes git diff to compare text between document versions.
[EXTERNAL_DOWNLOADS]: SKILL.md instructs the user or agent to install the global Node.js package docx (npm install -g docx) to enable document generation and formatting capabilities.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by ingesting and extracting data from untrusted .docx files.
Ingestion points: Document XML content is extracted in scripts/office/unpack.py and read by various processing and validation scripts.
Boundary markers: None identified; document content is processed without consistent delimiters to distinguish user data from agent instructions.
Capability inventory: The skill possesses significant capabilities for command execution and runtime code compilation, which could be abused if malicious instructions are successfully injected via the document processing pipeline.
Sanitization: While defusedxml is used in some components, other parts of the validation logic rely on standard xml.etree or lxml.etree without specific configurations to prevent XML-based exploits.

docx