The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains a hardcoded C source string that is written to a temporary file and compiled at runtime using gcc. The resulting shared object is then used via the LD_PRELOAD environment variable to intercept and modify system-level socket calls when executing LibreOffice (soffice). This is implemented as a workaround for environments with restricted UNIX socket permissions.
[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to execute system binaries for document processing and environmental setup.
Runtime compilation of shims using gcc in scripts/office/soffice.py.
Document conversion and rendering using soffice and pdftoppm in scripts/thumbnail.py and scripts/office/soffice.py.
Version comparison of document XML using git diff in scripts/office/validators/redlining.py.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied PowerPoint files, creating a surface for indirect prompt injection. Untrusted content from slides is rendered to images or extracted as text, which is then processed by the AI during analysis and visual QA tasks.
Ingestion points: User-provided .pptx files are ingested via scripts/thumbnail.py and scripts/office/unpack.py.
Boundary markers: The visual inspection prompts recommended in SKILL.md lack explicit boundary markers or instructions to ignore embedded commands in slide images.
Capability inventory: Subprocess execution for rendering and conversion; file-system access for unpacking and repacking XML structures.
Sanitization: Use of defusedxml.minidom in validation scripts helps mitigate XML-specific vulnerabilities like XXE, but does not filter natural language instructions embedded in slide content.

pptx