The Agent Skills Directory

[DYNAMIC_EXECUTION]: The toolkit's core workflow involves executing user-defined or agent-generated Python pipelines (pipeline.py). Additionally, it uses dynamic module loading in openmapstack/api.py to manage its library of semantic checks. These features are fundamental to the toolkit's design as an extensible analysis framework.
[COMMAND_EXECUTION]: The evaluation suite includes adapters (e.g., evals/adapters/openai_compatible.py) that drive AI models as autonomous agents by allowing them to execute shell commands and file operations within a sandboxed trial workspace.
[INDIRECT_PROMPT_INJECTION]: The skill has a large data ingestion surface, as it is designed to query and process geospatial data from external, untrusted sources such as OpenStreetMap (via Overpass), STAC catalogs, and various OGC services.
Ingestion points: openmapstack/connectors/ (DuckDB and PostGIS backends) and references/data-sources.md.
Boundary markers: The skill uses a structured project manifest (project.yaml) and explicit boundary definitions between source data, overrides, and processing logic.
Capability inventory: The toolkit possesses file-write capabilities (to data/derived/), subprocess execution (running the analysis pipeline), and network access (for data discovery and retrieval).
Sanitization: The tool implements read-only session enforcement for database connectors and performs automated credential scanning to prevent secrets from being recorded in the project manifest.
[EXTERNAL_DOWNLOADS]: The skill instructs agents to fetch geospatial data from official repositories, S3 buckets, and OGC APIs. It also specifies standard Python dependencies (e.g., duckdb, geopandas) and Node.js components for its distribution and evaluation framework. These downloads are from well-known technology providers and official data portals.

open-map-stack