The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py writes C source code to a temporary file at runtime and compiles it using gcc into a shared library (.so).
[DYNAMIC_EXECUTION]: The compiled shared library is injected into the LibreOffice process using the LD_PRELOAD environment variable. This is used to shim socket-related system calls (socket, listen, accept) to bypass environment restrictions in sandboxed containers.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to parse and extract text from external Word documents using tools like pandoc. This content is then provided to the agent's context, potentially allowing instructions within a document to influence agent behavior.
Ingestion points: SKILL.md (Read/analyze content tasks).
Boundary markers: Instructions do not explicitly specify the use of delimiters when reading external content.
Capability inventory: The skill can execute shell commands (soffice, pandoc, git) and perform file system operations (via scripts/accept_changes.py, office/soffice.py, office/validators/redlining.py).
Sanitization: The scripts use defusedxml for XML parsing which prevents XML External Entity (XXE) attacks, but does not sanitize potential prompt injection content within the documents themselves.

docx