Andersonlimahw Prompts — Data & Backend
Recipes for the data layer and managed-backend operations.
R1 — Well-modeled Supabase
Skills: supabase + supabase-postgres-best-practices
When: a new or reworked Supabase schema; RLS, indexes, and Postgres modeling.
Model <domain> in Supabase with the supabase skill, applying
supabase-postgres-best-practices: correct types, constraints,
justified indexes, RLS per role <roles>.
Success: ready migration SQL + RLS policy tested with one allowed
and one denied case per role; EXPLAIN on the main queries shows no
full scan.
R2 — Firebase operations
Skills: firebase-expert
When: Firestore/Auth/Functions/Hosting — modeling, rules, logs, or deploy.
Use firebase-expert on <project>: <task — e.g. review Firestore
security rules for <collections>, diagnose function <name> from
logs, structure collection <data> for read pattern <pattern>>.
Success: change applied with rules/queries validated (simulator or
emulator) and evidence of clean logs.
R3 — Continuous database monitoring
Skills: db-advisor
When: the database grows with nobody watching; you want recurring health reports, not discovering the problem during an incident.
Set up db-advisor to monitor <database/project>: recurring health
reports covering degrading queries, unused indexes, growth of
tables <tables>, and connections.
Success: first report generated with ≥1 prioritized action per
section; cadence defined and scheduled.
R4 — Production incident
Skills: incident-runbook + incident-center
When: something broke (or nearly did); response now, prevention after.
Incident: <symptom/alert>. Follow incident-runbook: triage,
mitigation, communication, evidence. Then log with incident-center
the blameless postmortem with preventive actions.
Success: mitigation applied with a recorded timeline; postmortem
with ≥2 preventive actions, each with an owner and a deadline.
See also
db-index-advisor— one-shot index fix per specific query (track prompts-engineering-quality R3); withdb-advisorthe hub-wide rule applies: one-shot × continuous, one driver per pass.api-test-loop— validate the API exposing this data.load-test/chaos-test— prove the data layer can survive production.