What problem does it solve? Supabase projects suffer from schema drift when new columns are added without guards, edge functions crash on PostgREST schema-cache errors (PGRST204/PGRST205), and diagnostic fields pollute the database schema. This Skill enforces a strict classification of stable DB columns versus JSONB debug fields so migrations and edge function writes never cause missing-column 400s or silently dropped diagnostics. ## Core Features & Use Cases - Field Classification Enforcement: Decides whether each new field becomes a stable DB column (queried, indexed, RLS-referenced) or lives inside geometry_report_json / source_context as JSONB. - Migration Guardrails: Requires IF NOT EXISTS guards, RLS impact notes, and NOTIFY pgrst reload statements on every migration touching roof_measurements, ai_measurement_jobs, or measurement_jobs. - Safe Edge Function Writes: Mandates strip-and-retry logic for PostgREST schema-cache errors, preserving removed fields in geometry_report_json.schema_drift_stripped_columns. - Result State Normalization: Locks result_state to a 10-bucket enum and routes specific solver failures to hard_fail_reason, failure_stage, and failure_details instead. - Use Case: When adding a new diagnostic payload to the AI measurement pipeline, the Skill tells you to nest it under geometry_report_json rather than creating a new column, and verifies the edge function write path has schema-cache retry handling. ## Quick Start Review my new Supabase migration and edge function insert for the roof_measurements table and confirm it follows the schema drift guard rules.
