The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted geospatial data (such as GeoParquet, Shapefiles, and GeoJSON) using the gpio CLI and DuckDB. This activity creates a potential surface for indirect prompt injection if the processed data contains malicious instructions intended to influence the agent's behavior.
Ingestion points: The skill utilizes gpio inspect, gpio convert, and gpio check on user-supplied files in SKILL.md and references/gpio-commands.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded content were found in the skill definitions.
Capability inventory: The skill has the ability to write files to the local system and upload data to remote cloud storage via gpio publish upload.
Sanitization: There are no specific sanitization or validation routines mentioned for the content of the geospatial files beyond format checking.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external Python packages, including geoparquet-io, duckdb, and the gpio-pmtiles plugin. duckdb is a well-known spatial database library, while geoparquet-io and its plugins are the primary tools provided by the vendor for these workflows.
[COMMAND_EXECUTION]: The skill relies on the execution of the gpio command-line interface for various tasks such as sorting, partitioning, and metadata inspection. It also utilizes the DuckDB CLI and potentially tippecanoe for vector tile generation.
[DATA_EXFILTRATION]: The skill provides functionality to upload processed GeoParquet files to external destinations, specifically Amazon S3 buckets, using the gpio publish upload command. This is documented as a standard feature for distributing optimized spatial data.

geoparquet-validation