The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines processes for ingesting data from external and untrusted sources, creating a potential surface for indirect prompt injection.
Ingestion points: The skill uses gpd.read_file() and gpd.read_postgis() in SKILL.md and references/data-io.md to load data from external URLs and databases.
Boundary markers: Absent. No instructions are provided to treat external data as untrusted.
Capability inventory: The skill possesses capabilities for file system writes, network access, and database interactions.
Sanitization: No sanitization logic is specified for data processed from external sources.
[EXTERNAL_DOWNLOADS]: The skill demonstrates the retrieval of geospatial data from external network locations.
Evidence: references/data-io.md contains examples of fetching files via HTTP/HTTPS (https://example.com/data.geojson), Amazon S3 (s3://bucket/data.gpkg), and Azure Blob Storage (az://container/data.gpkg).
[CREDENTIALS_UNSAFE]: The documentation includes a database connection string with placeholder credentials.
Evidence: references/data-io.md contains the example string postgresql://user:password@host:port/database.
[PROMPT_INJECTION]: The skill metadata lists the author as 'K-Dense Inc.', which differs from the provided context author 'qinyan-ai'.

geopandas