AI SQL & Database Prompt Pack

AI writes SQL in seconds. The trouble is that wrong SQL rarely fails loudly. It returns a number, and the number looks plausible.

Ask ChatGPT, Claude, Cursor, Claude Code or Codex for "revenue per customer last month" and you'll get a query that runs. Then you look closer. It joins invoices to line items and sums the invoice total once per line. It uses date_trunc on MySQL. It filters with NOT IN over a column that contains NULLs, so it quietly returns nothing for one segment. It invents users.signup_at because that's what the column is usually called. And the "add an index on every WHERE column" advice ignores the plan entirely.

AI SQL & Database Prompt Pack gives you 32 ready-to-paste prompts that make the model work from your real schema, your engine and version, and your business definitions, state its assumptions, explain join cardinality, ground performance claims in EXPLAIN output, and hand you a way to verify every answer: a tiny reproduction, a reconciliation query, a test.

What you get

• 📘 PDF guide (35 pages): why AI-written SQL goes wrong, the database context block (schema, dialect, constraints), all 32 prompts with when-to-use notes and what to watch for, combinations, an illustrative (fictional) worked example, prompt anatomy and follow-up questions
• 🗂️ templates/: 32 plain-Markdown prompts + 3 checklists + README, ready to copy into your repo or notes

The 32 prompts

• Writing SQL (01–04): reusable schema context block · natural language → SQL with grain, assumptions and sanity checks · clarify a vague request before writing · port queries between PostgreSQL, MySQL and SQLite without changing meaning
• Query review (05–10): correctness review (9 checks, verdict per item) · NULL handling audit · join fan-out / double counting · N+1 hunt across ORMs · sargability and index use · pagination (OFFSET vs keyset)
• Query plans (11–14): PostgreSQL EXPLAIN (ANALYZE, BUFFERS) · MySQL EXPLAIN / FORMAT=TREE / EXPLAIN ANALYZE · SQLite EXPLAIN QUERY PLAN · before/after plan and result comparison
• Schema & indexing (15–18): schema design review · normalization and redundancy · indexing strategy from real queries · move integrity rules from app code into constraints
• Migrations (19–22): migration safety review (locks, rewrites, data loss, old-code compatibility) · zero-downtime expand → migrate → contract plan · batched, idempotent backfills · rollback plans
• ORM & injection (23–27): Prisma · SQLAlchemy · Django ORM · ActiveRecord · raw-SQL injection check for any stack
• Test data & tests (28–29): deterministic seed data with real edge cases · database tests on the same engine as production
• Reporting (30–32): window functions · multi-step CTE reports (cohorts, funnels, retention) · verify a report before anyone acts on it

Also included

• Before you run AI-written SQL: a verification checklist (review every statement, never run AI DDL/DML on production first, transactions, read-only roles, timeouts, backups, red flags)
• Migration preflight checklist: actual SQL, lock analysis, online index builds, compatibility, backfills, rollback
• Engine quick reference: PostgreSQL vs MySQL vs SQLite differences that most often break AI-written SQL (EXPLAIN syntax, upserts, NULL-safe equality, integer division, transactional DDL, partial indexes, FK enforcement and more), with version notes

Who it's for

• Indie devs and small teams who ship with AI coding tools and touch the database themselves
• Backend and full-stack developers who want a second pair of eyes on queries, plans and migrations
• Anyone using Prisma, SQLAlchemy, Django or Rails who has been surprised by the SQL their ORM produced

Who it's not for

• DBAs looking for deep engine internals or server tuning guides
• Anyone hoping to let an AI agent run migrations on production unattended. This pack is built around human review

How it works

1. Build your database context block once with prompt 01 and save it in your repo
2. Pick the prompt for the job and fill in the {{placeholders}}
3. Read the verdicts, assumptions and evidence before running anything
4. Verify with the checklist, a reconciliation query or a test; change schemas with a phased plan and a rollback

Honest positioning

Engine notes in the guide (October 2026) include version caveats and point to the official PostgreSQL, MySQL and SQLite docs; confirm details for your exact version. No fabricated testimonials, download counts, speed-up figures, time-saved claims or benchmarks. The worked example is labelled as fictional. Judge it by running prompt 05 or 07 on a query you already trust and seeing what it finds.

Made by Pluto (X @Mars_Peakul).