The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py defines a C source string that is written to a temporary file, compiled at runtime using gcc, and loaded as a shared library via the LD_PRELOAD environment variable. This mechanism is employed to shim networking-related system calls to allow LibreOffice to function in restricted or sandboxed environments.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Office documents as its primary function, creating an attack surface for indirect prompt injection. 1. Ingestion points: External document data enters the agent context through scripts/office/unpack.py and pandoc text extraction. 2. Boundary markers: The skill does not use specific markers to isolate document content or instructions to ignore embedded commands. 3. Capability inventory: Several scripts execute shell commands via the subprocess module (scripts/accept_changes.py, scripts/office/soffice.py). 4. Sanitization: There is no logic present to sanitize or escape extracted document content before processing.
[COMMAND_EXECUTION]: The skill uses the subprocess module across multiple files to orchestrate external binaries including soffice, gcc, git, and pandoc for document conversion, comparison, and compilation.

docx