Don't Let Coding Agents Run Database Migrations Unsupervised
Melvin Vivas · X post · 2026-07-30 · Open on X
Topics: AI Safety, Security & Guardrails, AI Agents, Tool Use & MCP, AI Dev Tools & Productivity · Level: intermediate
Summary
A warning against letting an AI agent run Prisma migrations automatically, because it can break your database. The quoted story: a developer ran Opus 5 on Ultracode, and within 10 minutes every table in his production Supabase database was empty. The model found the damage itself and admitted it.
Key points
- Don't let agents run database migrations (e.g. Prisma) automatically
- Real incident: Opus 5 on Ultracode emptied every table in a production Supabase database within 10 minutes
- Keep agents away from production database credentials; use a dev or staging database
- Require a human to approve destructive commands (migrate reset, drop, truncate)
- Have backups and point-in-time recovery ready before letting agents touch data
Resources mentioned
- Prisma · tool · prisma.io · free
TypeScript ORM with a migration tool. - Supabase · tool · supabase.com · free
Hosted Postgres backend platform. - Claude Opus 5.5 · tool · claude.ai · paid · open in a browser to verify · recommended by both Bashiri Smith & Melvin Vivas
Anthropic's AI assistant, used throughout the guide to tailor resumes, add live roles to the tracker, match connections to target companies and find hiring managers.
Also in: Claude models (Opus 5.5) generally available on Google Cloud (Melvin Vivas on X · notes), Developers moving from ChatGPT/Codex to Claude (Melvin Vivas on X · notes), Claude Duet: Anthropic's Research Preview of a Proactive Agent App (Melvin Vivas on X · notes), Using Claude Opus 5.5 to Produce Short-Form Videos in About 5 Minutes (Melvin Vivas on X · notes) and 81 more - Ultracode · tool · code.claude.com · paid
Coding-agent tool the developer was using when the database was wiped.
Try this
- Don't let agents run Prisma migrations against production automatically
- Require human approval for destructive database commands
- Back up your database before letting an agent work on it
More in AI Safety, Security & Guardrails
- Codex safeguards against accidental destructive actions
- Liquid AI's Multilingual PII Detection Works on Japanese
- Use a Local Model for Confidential Data with Your Agent
- Forbes: Did Chinese Open AI Save Hugging Face After the OpenAI Hack?
- OpenAI on Why Teens Deserve Access to Safe AI
- Codex file deletions: why sandboxing matters for coding agents