The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides several Python CLI utilities (e.g., image_metadata.py, palette_audit.py, export_plan.py) for auditing scientific visualizations. These scripts are executed using the uv tool with pinned dependencies and are designed for deterministic, network-free operation.
[EXTERNAL_DOWNLOADS]: Dependencies are limited to well-known scientific libraries (Matplotlib, Seaborn, Plotly, Pillow, pypdf) from the official Python Package Index (PyPI). The skill also includes instructions to fetch citation metadata from the reputable arXiv repository.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data (tabular data and images). It mitigates injection and denial-of-service risks by implementing strict path validation (rejecting symlinks), enforcing input size limits, and using defensive parsing for SVG files to block DTD/entity declarations.
[SAFE]: The implementation follows security best practices, such as atomic file writing through temporary files, explicit permission management (0o600 for temporary artifacts), and the use of pixel-count limits in image processing to prevent decompression bomb vulnerabilities.

scientific-visualization