The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through the processing of complex external files.
Ingestion points: Reads and extracts content from user-provided PowerPoint (.pptx) files using ooxml/scripts/unpack.py and processes HTML content for slide generation in scripts/html2pptx.js.
Boundary markers: Absent when document content is extracted and presented to the agent for editing or analysis.
Capability inventory: The skill has permission to use the Bash tool and performs multiple system calls via subprocess.run, providing an exploitation path if malicious instructions are processed.
Sanitization: While defusedxml is used for XML parsing, the zip extraction logic in ooxml/scripts/unpack.py and ooxml/scripts/validation/base.py uses zipfile.extractall() without member validation, making the environment susceptible to Zip Slip (arbitrary file write) attacks.
[COMMAND_EXECUTION]: The skill executes various system utilities to perform document conversion and validation tasks.
Evidence: Uses subprocess.run to invoke soffice (LibreOffice) for PDF/HTML conversion, pdftoppm (Poppler) for thumbnail creation, and git diff for change validation in ooxml/scripts/pack.py, scripts/thumbnail.py, and ooxml/scripts/validation/redlining.py.

pptx