The Agent Skills Directory

[SAFE]: The skill provides legitimate instructions for using standard geospatial tools such as Tippecanoe, PMTiles, and GeoPandas. No malicious behavior, unauthorized credential access, or suspicious remote code executions were detected.
[INDIRECT_PROMPT_INJECTION]: The skill documents a pipeline that ingests external geospatial data and executes shell commands, creating a potential surface for indirect prompt injection if the agent processes untrusted inputs.
Ingestion points: The pipeline processes external data files including raster.zarr, geoparquet shards, and FlatGeobuf batches.
Boundary markers: The instructions do not define explicit delimiters or warnings for the agent to ignore instructions that might be embedded in the metadata of the geospatial files.
Capability inventory: The skill uses Python scripts (via libraries like GeoPandas and Pyogrio) and shell-based tools (tippecanoe, tile-join, go-pmtiles) which are invoked with user-supplied or data-derived parameters.
Sanitization: The pipeline includes data-level sanitization such as dropping NULL, empty, or invalid geometries (geom.is_valid), which mitigates technical errors but does not specifically sanitize against instruction injection in string attributes.

pmtiles-pipeline