The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py dynamically generates and executes code. It writes a C source file to a temporary directory and invokes gcc to compile it into a shared object library. This library is then loaded using LD_PRELOAD to modify system-level socket behavior, which is a high-risk capability even if used for environment compatibility.
[COMMAND_EXECUTION]: The skill frequently invokes external CLI tools using subprocess.run. Affected files include scripts/accept_changes.py, scripts/office/soffice.py, and scripts/office/validators/redlining.py, which execute soffice (LibreOffice), gcc, and git.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting and processing untrusted document content.
Ingestion points: scripts/office/unpack.py extracts XML from arbitrary .docx files into the local environment.
Boundary markers: Absent; XML content is processed and merged without delimiters or instructions to ignore embedded prompts.
Capability inventory: The skill has extensive command execution capabilities (soffice, gcc) that could be targeted via document content poisoning.
Sanitization: Absent; there is no evidence of content filtering or sanitization before processing the XML files.

docx