The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various system commands to perform document processing tasks. \n
scripts/office/soffice.py runs gcc to compile a shim library and then executes the soffice binary. \n
scripts/accept_changes.py invokes soffice in headless mode to process tracked changes in Word documents. \n
scripts/office/validators/redlining.py executes git diff to compare and validate text content differences between document versions.\n- [REMOTE_CODE_EXECUTION]: The skill performs dynamic code compilation and library injection. \n
scripts/office/soffice.py writes a C source string to a temporary file, compiles it using gcc, and then uses the LD_PRELOAD environment variable to inject the resulting shared library into the LibreOffice process. This technique is used to bypass network socket restrictions in sandboxed environments but involves high-risk execution patterns. \n
scripts/accept_changes.py generates and writes a LibreOffice Basic macro to a configuration file before executing it, representing dynamic script generation.\n- [PROMPT_INJECTION]: The skill extracts content from untrusted document files, exposing an indirect prompt injection surface. \n
Ingestion point: Document XML content is unpacked and extracted in scripts/office/unpack.py. \n
Boundary markers: The skill's instructions do not include specific delimiters or warnings for the agent to ignore potentially malicious instructions found within extracted document text. \n
Capability inventory: The skill provides the agent with powerful tools for command execution and file system access across multiple scripts. \n
Sanitization: While defusedxml is used to mitigate XML-specific attacks, there is no validation or filtering of the natural language text content extracted from the documents.

docx