geobase-project-db-data-import
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external local files and user-defined table names, creating an ingestion surface for potentially untrusted data that could influence agent behavior.
- Ingestion points: Local geospatial files (GeoParquet, GeoJSON, GPKG, Shapefile, CSV, Parquet) and target table names.
- Boundary markers: The skill employs shell quoting for file paths and uses PostgreSQL identifier formatting (
%I) for table existence checks. - Capability inventory: The skill is capable of executing shell commands (
ogr2ogr,geobase-cli,duckdb), reading local files, and executing SQL queries. - Sanitization: Explicit instructions are provided to sanitize table names (allowing only alphanumeric characters and underscores, with a 63-character limit) and to verify table availability before proceeding.
- [COMMAND_EXECUTION]: The skill relies on shell commands to perform data inspection and ingestion tasks.
- Evidence: Usage of
ogr2ogrfor database imports andgeobase-clifor environment management. - Safety measures: Database credentials like
GEOBASE_PGPASSWORDare handled via environment variables rather than command-line arguments, and the skill explicitly forbids the agent from requesting secrets within the chat context.
Audit Metadata