open-map-stack
Use textual agent instructions for GIS and geospatial work: source discovery and provenance, vector/raster/point-cloud pipelines, CRS and metric analysis, spatial SQL, routing and isochrones, QGIS projects, tile generation, and web maps. Use advanced tools and formats such as OSM, Overture, STAC, Sentinel/Landsat, LiDAR, GeoPackage, GeoParquet, COG, PMTiles, WMS/WFS/OGC APIs, GDAL, GeoPandas, DuckDB Spatial, PostGIS, QGIS, MapLibre, and Estonian spatial data including ETAK and EPSG:3301. Open-first, with hosted services when scale or reliability requires them. Do not use for casual map references, simple place lookups, or ordinary travel directions without analytical GIS work.
jaakla · AI review on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add jaakla/openmapstack --skill open-map-stackAbout this skill
name: open-map-stack description: "Use textual agent instructions for GIS and geospatial work: source discovery and provenance, vector/raster/point-cloud pipelines, CRS and metric analysis, spatial SQL, routing and isochrones, QGIS projects, tile generation, and web maps. Use advanced tools and formats such as OSM, Overture, STAC, Sentinel/Landsat, LiDAR, GeoPackage, GeoParquet, COG, PMTiles, WMS/WFS/OGC APIs, GDAL, GeoPandas, DuckDB Spatial, PostGIS, QGIS, MapLibre, and Estonian spatial data including ETAK and EPSG:3301. Open-first, with hosted services when scale or reliability requires them. Do not use for casual map references, simple place lookups, or ordinary travel directions without analytical GIS work."
OpenMapStack Toolkit
Production-grade geospatial workflows with an open-first stack and pragmatic hosted/SaaS choices when global scale, latency, SLA, or data quality makes local processing a poor fit. Cloud-native by default: STAC for discovery, GeoParquet + COG + PMTiles for storage, DuckDB and PostGIS for compute, MapLibre and Martin for delivery.
Reproducible project-first contract
Core principle: reasoning may be exploratory; the delivered analysis must be deterministic, inspectable, and reproducible.
For any material multi-stage GIS analysis, do not optimize for reaching the final map, dashboard, or answer quickly. A one-off polished dashboard is not the deliverable — a reproducible technical GIS project is. First establish a reusable project artifact, then derive the map/dashboard/report from it.
Before treating an analysis as complete, you MUST compile (or maintain) a project like examples/tartu-development: a canonical project.yaml (openmapstack-project/v1), pipeline.py, and README.md; pinned sources with timestamps, selections and licensing; explicit assumptions; every manual addition or correction stored as real geodata; deterministic ordered steps with explicit CRS; machine-readable validation rules plus the report from the run; and output definitions with semantic presentation intent and provenance surfaced in the rendered view. references/project-spec.md is the full schema — read it before compiling a project.
Workflow (agent may retry/experiment internally, but the accepted analysis is recompiled deterministically):
USER QUESTION
↓
interpretation / exploration (internal, may be ad-hoc)
↓
COMPILE GIS PROJECT project.yaml + pipeline + manifest + overrides + validation
↓
EXECUTE PROJECT
↓
validated derived datasets
↓
QGIS project / standardized web view
↓
FINAL ANALYSIS / DASHBOARD / ANSWER
The polished map/dashboard is a view over the project, not the canonical definition of the analysis.
Hard rules for every material analysis — each is expanded in references/project-spec.md:
- Real source data mandatory; never hallucinate coordinates. Fabricating coordinates or synthesizing baseline geometry is forbidden without explicit, informed user consent. Hypothetical or planned features go in
data/overrides/with provenance, rationale, and evidence. - Never mutate source data. Immutable source + project override layer = effective input. Distinguish external facts, transformations, corrections, assumptions, and hypothetical data.
- Overrides must be executable and verified. Target a real source feature; for attribute changes the asserted prior value must match. Evidence must be non-placeholder. Validation reports each override
applied,rejected, ornot_testable— listing one is not applying it. Scenario features stay labeled hypothetical and visually distinct from authoritative layers. - Record, don't memoize on chat. Encode every manual fix as data or pipeline logic. A fresh environment with the documented sources must reproduce the project; the transcript is not part of the dependency graph.
- Prove semantic predicates from data. Ownership, active status, public access, legal designation: the source must expose an authoritative field or documented mapping. Preserve unknown as unknown; never default a missing value to the desired class.
- Bounded APIs must prove completeness. Record
numberMatched/equivalent and page until returned == matched. A response filled to the request limit is incomplete until proven otherwise. - Validation is a pipeline stage, not prose advice, with machine-readable results. Every declared check appears exactly once in the report;
warning/not_testablepropagate to run and project status; run IDs and hashes resolve to a realruns/*.jsonrecord. - Run the project CLI when available. Use
openmapstack validate project.yamlbefore delivery andopenmapstack run project.yamlfor the canonical execution path. The CLI audits the manifest, provenance, graph, artifacts, report, and run record; it does not replace domain GIS checks performed by the pipeline. - The manifest must resolve. Every step input is a source key or an earlier step's output, spelled as the producer declared it; every
generated_bynames a real step (manifest_graph_resolves). - One canonical implementation creates every declared output. Convenience/E2E entrypoints may wrap
pipeline.pybut must not duplicate its processing, QGIS, or report logic. - Build a layer- and style-perfect QGIS project (
project.qgz) mirroring the web view: matching layer-tree groups, identical categorized styles,./path.gpkg|layername=namedatasources, and a regional tiled basemap. Success means valid layers, not exit code 0 — pin the runtime, and when PyQGIS is available require every layerisValid(); otherwise recordnot_testable, never an implicit pass. Two traps make a project that passes every one of those checks still show the wrong map, so check them explicitly:- Every layer declares its own
<srs>, basemaps included. A layer without one is assumed to be in the project CRS and never reprojected — a Web Mercator bas
- Every layer declares its own