The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches authoritative geospatial datasets from well-known regional services and government infrastructure.
Example: Downloads cadastral data from Estonian Land Board S3 storage (s3.pilw.io).
Example: Queries the Environment Agency's GeoServer (gsavalik.envir.ee) for national road network data via WFS.
[DYNAMIC_EXECUTION]: The skill generates and uses Python processing pipelines to ensure analysis is reproducible.
The templates/pipeline.py file uses an inline __import__("json") call for report serialization, which is a standard development practice for keeping template scripts portable.
The skill encourages the agent to generate code (pipeline.py) to perform multi-stage analysis, which is fundamental to its geospatial purpose.
[COMMAND_EXECUTION]: The example pipeline in examples/tartu-development/pipeline.py includes functionality to execute QGIS within a Docker container to compile project files natively.
This execution is scoped to local workspace paths and is intended for high-fidelity project generation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external APIs (such as WFS and ArcGIS REST services).
Ingestion points: External data is ingested via urllib.request and processed by duckdb and geopandas in pipeline.py scripts.
Boundary markers: Not explicitly defined in generated scripts, but the skill instructions emphasize using real, verified source data.
Capability inventory: The environment includes powerful GIS tools, SQL execution via DuckDB, and file-writing capabilities.
Sanitization: Standard GIS format parsers (GDAL/OGR) provide inherent structure, though they do not provide specific prompt-injection sanitization.
Severity: Low, as the ingested data consists of structured geospatial records from authoritative providers.

open-gis