2.7 KiB
2.7 KiB
CODEX FIRST DAY MASTER PROMPT — M14 BUILD LAUNCH
You are implementing GeoIntel Kempen.
GeoIntel is a GeoAI Workbench for the Belgian Kempen. It is not a generic dashboard, chatbot, QGIS clone or report-only app. The product kernel is:
data -> processing -> geospatial output -> QA/QC -> export
Mandatory read order before coding
CODEX_START.mddocs/00-start/START_HERE.mddocs/governance/GEOINTEL_CONSTITUTION.mddocs/governance/ARCHITECTURE_INVARIANTS.mddocs/governance/FORBIDDEN_DECISIONS.mddocs/40-build-launch/SPRINT_1_SCOPE_FREEZE.mddocs/40-build-launch/BUILD_SUCCESS_DEFINITION.mddocs/40-build-launch/BUILD_ORDER_GRAPH.mddocs/40-build-launch/CODEX_STOP_RULES.mddocs/40-build-launch/MODULE_ACCEPTANCE_CRITERIA.mddocs/specs/GIS_STANDARDS.mddocs/specs/STATE_MACHINES.mddocs/30-codex-optimization/PROMPT_DISCIPLINE.md
First implementation mission
Build Sprint 1 only.
Sprint 1 goal:
Project -> Area -> Dataset upload/registration -> Metadata extraction -> PostGIS persistence -> Minimal map display -> GeoJSON export baseline
Build exactly this first
- Backend foundation.
- Database/PostGIS foundation.
- Project API.
- Area API.
- Dataset API for GeoJSON.
- Storage service.
- Metadata extraction for GeoJSON.
- Frontend shell.
- MapLibre map.
- Minimal dataset/area UI.
- Readiness/smoke tests.
Do not build yet
- YOLO live inference.
- SAM live segmentation.
- GRB live WFS.
- Sentinel automation.
- LiDAR.
- Training Studio.
- AI Copilot.
- Full report generator.
- Multi-user auth.
If you encounter a task that belongs to a later module, add it to backlog/open issues and continue Sprint 1.
Quality rules
- Do not introduce undocumented technologies.
- Do not commit secrets.
- Do not hardcode machine-specific paths.
- Do not use permanent mock data.
- Golden fixtures are allowed for tests and initial UI development.
- Keep API responses canonical.
- Keep geometry rules consistent with GIS standards.
- Add tests/smoke checks where practical.
- Update docs/status/changelog when behavior changes.
Stop rules
Stop and report instead of continuing if:
- build breaks;
- tests fail and cannot be fixed within the active scope;
- architecture invariant would be violated;
- API contract conflict appears;
- a new dependency seems required but is not in ADRs;
- Sprint 1 scope would be exceeded.
Required final response after implementation pass
Report:
- files changed;
- implementation summary;
- commands run;
- test results;
- known limitations;
- architecture invariants touched or confirmed untouched;
- out-of-scope items deferred;
- next recommended pass.