The Agent Skills Directory

Dynamic Code Execution: The skill supports the execution of author-supplied Python scripts (content-extra.py) to handle complex drawing logic. This capability is managed through several security layers:
Execution is disabled by default and requires the explicit --allow-scripts flag from the operator.
Before execution, the skill performs AST-based static analysis to block dangerous modules (e.g., os, subprocess, socket) and builtins (e.g., eval, exec).
The documentation explicitly notes that the linting is a defense-in-depth measure rather than a hardened sandbox.
External Downloads: The skill's documentation and orchestrator scripts provide commands to download and install the uv tool from its official source (astral.sh). This is a well-known service used for Python package management.
Command Execution: The skill invokes LibreOffice (soffice) via subprocess.run to perform PPTX-to-PDF conversions. These invocations are implemented using argument lists rather than shell strings, which is a secure pattern that prevents shell command injection.
Binary File Safety: The skill includes a dedicated pdf_safety.py module that implements defense-in-depth checks for PDF parsing. It validates file regularity, size ceilings, magic-byte prefixes, and page counts before passing data to the underlying C-based parser, reducing the risk from malformed or malicious documents.
Indirect Prompt Injection Surface: The validate_slides.py script sends slide content to vision models for quality inspection. This process involves processing potentially untrusted text and images within the model's context. The risk is minimized by the skill's specific focus on structural and visual quality checks.

powerpoint