Tool calling
How GIS skills guide Hermes to use the tools registered in the active profile.
Skills and tool calls
A selected skill contributes workflow rules, quality checks, and safety boundaries to the model context. When the active profile exposes a compatible tool, Hermes can then call it using that tool's registered schema.
request
→ load arcgis-rest-service-validator
→ inspect the connected tool schemas
→ call a public HTTP or ArcGIS metadata endpoint
→ validate count, paging, geometry, CRS, and fields
→ return evidence and limitationsTool schemas come from the active profile
Hermes declares enabled tools to the model with their names, descriptions, and argument schemas. The model can call only the tools that are actually registered for that session. Results return as structured tool output before the final answer is written.
The GIS skill package does not register a GIS toolset. It adds procedural knowledge that can guide built-in web, browser, terminal, and file tools, or separately configured MCP servers and custom integrations.
Verify names and schemas before publishing examples
Do not copy a tool-call example until the tool name and JSON schema have been
verified in the current Hermes installation. The skills repository does not
register esri.query_features, open-data.search, or
postgis.run_readonly_sql by itself.
A safe spatial lookup sequence
An address-to-zoning lookup is usually at least two operations:
- Geocode the approved address to point geometry.
- Query the zoning layer with that geometry and an ArcGIS spatial relationship.
- Request only necessary output fields.
- Validate result count, source, and spatial reference.
- Report the district code, source URL, timestamp, and caveats.
The exact calls depend on the registered integration. If no compatible tool is available, the skill can still produce a validation plan and implementation commands without pretending execution occurred.
Write operations
For publishing, edits, database writes, or hosted-content changes, tool implementations should enforce least privilege, exact target verification, preview or diff output, approval tied to the final arguments, and a rollback path. Skill text is guidance; it is not a hard security boundary.