The Agent Skills Directory

[SAFE]: The skill exhibits an exceptional security posture by defining a comprehensive 'Safety and privacy contract' in SKILL.md. This contract mandates local-only operations, prohibits automatic loading of URLs or archives, and enforces the use of secret managers for database credentials.
[SAFE]: The shared utility module _common.py implements rigorous input validation via reject_nonlocal and _candidate_path functions. These functions effectively block path traversal, remote URI schemes (e.g., http, s3, vsi), and interaction with symbolic links or compressed archives.
[SAFE]: The bundled CLI scripts (e.g., vector_inventory.py, geometry_validity_report.py) are designed with 'privacy-by-default' principles. They extract technical metadata and aggregate statistics while explicitly redacting exact coordinates, feature identifiers, and detailed bounding boxes to prevent data exposure.
[SAFE]: Resource consumption is strictly controlled through hard ceilings on input file sizes (ABSOLUTE_MAX_INPUT_BYTES), feature counts (ABSOLUTE_MAX_FEATURES), and join multiplicities, protecting the agent environment from resource exhaustion or 'billion laughs' style attacks on geospatial parsers.
[SAFE]: External dependencies are pinned to specific versions of reputable libraries (e.g., GeoPandas, Shapely, PyProj) from standard package registries, minimizing supply chain risks.
[SAFE]: The skill provides clear guidance on secure PostGIS connectivity, demonstrating how to use SQLAlchemy with environment-based credentials while warning against the exposure of connection strings in logs or source code.

geopandas