Reproducible GIS project

Turn the analysis into an inspectable project another analyst can execute without the conversation. Retain the user's chosen algorithm and stack unless they cannot meet a stated requirement; explain consequential changes before making them. Do not silently substitute a convenient proxy for the requested measurement.

When the user has dlt, Dagster, dbt, QGIS or Observable in their stack, read references/established-stack.md for ownership and integration boundaries. Preserve their chosen tools; retain source pins, GIS correctness and reproducible evidence. Do not duplicate their SQL, ingestion or scheduler definitions.

Compile or update the project

Read references/project-workflow.md for mandatory workflow and delivery requirements and references/project-spec.md for the project contract. Use the local templates/ and examples/tartu-development/ as structural examples, not as evidence about the user's data. references/installation.md explains CLI setup and the worked example's runtime requirements.

1. Establish the analysis objective, authoritative inputs, semantic filters, metric CRS, algorithm, outputs and consequential unresolved assumptions. Inspect an existing project before changing its structure. If architecture is still undecided, use an installed open-map-stack or resolve those decisions with the user before presenting the project as complete.
2. Pin immutable sources and represent corrections/scenarios as explicit data with provenance. Verify each correction's target and prior value before applying it. Never fabricate missing authoritative geometry or overwrite a source to make validation pass.
3. Declare one canonical executable pipeline and ordered, resolvable steps. Record delivery.schema: openmapstack-delivery/v1 for new projects; use the built-in dashboard by default, or the user's explicit external-only/combined selection. QGIS and Observable are optional targets. Existing manifests without delivery retain legacy v1 behavior until explicitly migrated. Derive selected views from that pipeline and manifest; avoid independent presentation logic that silently changes the analysis.
4. Execute relevant checks, retain machine-readable results and run evidence, and perform a clean rerun. Check whether the openmapstack CLI is installed (openmapstack --version or python3 -m openmapstack --version, then use whichever form works for every command); if it is, run openmapstack validate, then openmapstack verify project.yaml --rerun for the clean rerun. Make that verify --rerun your last step and do not deliver until it passes; a passing validate alone is not enough. The canonical pipeline must produce every file under data/derived, every selected deliverable, the run record, and the manifest's runs.latest and project.status, never a one-off script or a hand edit: a rerun writes a new run record, so a hand-patched pointer breaks it. Missing tools or unknown data semantics remain visible limitations. A stated intention to validate is not validation.
5. Generate the findings artifact from pipeline results before output hashing. Declare the real author and toolkit, and render linked credits with the canonical run time in every delivered view (workflow credits requirement). Review the required attribution prose too; field checks cannot certify its meaning. Deliver the project, source/override policy, runtime instructions and substantive method limitations together. Preserve the full presentation obligations for the selected targets in the workflow reference; a view alone is not the project.

Focused supporting guidance

• CRS, formats and metric correctness: references/formats-and-crs.md.
• User warehouses and reproducible snapshots: references/user-data-sources.md.
• Authoritative public inputs and catalog completeness: references/data-sources.md.
• Validation and operational evidence: references/validation-and-ops.md.
• QGIS delivery and runtime behavior: references/qgis.md.
• Web basemap selection, global fallback and light/dark styles: references/web-delivery.md.
• Dashboards, reports and their checked interaction hooks: references/design-language.md.

An installed geospatial-data-discovery or spatial-sql may help with a bounded source or query question. Their absence does not prevent this skill from using its own references. Keep ownership of the compiled project and cross-stage correctness; do not require or automatically install sibling skills.

For optional product expertise, read references/companion-skills.md. Keep GIS correctness and the task boundaries above when using a companion.