The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/office/soffice.py dynamically generates C source code at runtime, writes it to a file, and invokes gcc to compile it into a shared object library. This library is then injected into the execution environment of the soffice process using the LD_PRELOAD environment variable to intercept system-level socket calls.
[COMMAND_EXECUTION]: The skill extensively uses the subprocess module to execute external binaries and system tools. Key findings include:
scripts/office/soffice.py executes gcc for binary compilation and soffice (LibreOffice) for document processing.
scripts/accept_changes.py executes soffice with headless flags and custom installation environments to apply document macros.
scripts/office/validators/redlining.py executes git to perform word-level comparisons between document versions.
[PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection by ingesting and processing untrusted document files.
Ingestion points: scripts/office/unpack.py and scripts/office/validate.py extract and analyze XML content from user-provided .docx files.
Boundary markers: Absent. The skill does not employ specific delimiters or instruction-ignore markers when processing external document text.
Capability inventory: The skill environment provides high-privilege capabilities including binary compilation, process injection, and shell access to tools like pandoc and soffice.
Sanitization: While the skill uses defusedxml to mitigate XML-specific vulnerabilities, it does not demonstrate sanitization of document content before passing it to system utilities.

docx