Supabase DB Supervisor — Data skill for Claude Code
Fresh-eyes supervision pass for AI-built database work.
How to install Supabase DB Supervisor
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open CarolMonroe22/supabase-db-supervisor and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Supabase DB Supervisor does
Fresh-eyes supervision pass for AI-built database work. The builder never grades its own homework.
Alternatives in Data
- Data Contract — This document defines which files belong to the system (auto-updatable) and which belong to the user (never to 32.5k ★
- MCP Server Supabase — Supabase database and auth integration 2.6k ★
- Pipeline Feature — Add a feature to the FFmpeg transcode pipeline following Tunarr's pipeline builder pattern 2.5k ★
README
supabase-db-supervisor
An agent skill that adds a **fresh-eyes supervision pass** to any database work an AI agent builds. A separate agent plays the investigator, proves who can see which data with real queries, and gives an explicit OK before anything ships. The builder never grades its own homework.
**Status: work in progress.** This is v0.1, a starting point published in the open. The checklist, routing logic, and output format will evolve. Issues and PRs welcome.
Why
[Supabase Evals](https://supabase.com/evals) (open sourced July 2026) measured how well coding agents build real Supabase backends across four stages. The headline:
| Stage | What it means | Top agents |
|---|---|---|
| Build | Create schema, auth, RLS from scratch | ~100% |
| Deploy | Ship it live with secrets and monitoring | ~100% |
| Investigate | Something broke, find out why | 67% for most |
| Resolve | Fix it without breaking more | ~100% |
Agents are excellent at the part where you feel productive, and weakest at the part where you feel lost. This skill moves the Investigate stage inside the build loop, so the gap gets covered before anything ships instead of at 11pm when it breaks.
How it works
flowchart LR
A[Agent builds
database work] --> B[Fresh-context
supervisor]
B --> C[Access proofs
real queries]
B --> D[Advisors
security + perf]
B --> E[Migrations
vs reality]
B --> F[Upstream
freshness]
C --> V{VERDICT}
D --> V
E --> V
F --> V
V -->|OK| S[Ship 🏁]
V -->|NEEDS ATTENTION| R[Fix +
Related Skills
Zju Scholar
ZJU Scholar — Claude Code Skill for Zhejiang University academic data (timetable, grades, exams, homework, Zhi
Multi Agent Customer Support
Multi-agent AI customer support system for a digital music store. Built with LangGraph, Gradio, and the Chinoo
Hunter
You are Hunter in an automated bug-fixing pipeline. You work in a fresh clone of the repository on a build mac
External Website Builder To Production
A Claude Code skill that takes a website built somewhere else (Webflow, Framer, Readdy, custom code, etc.) and
Pioneer
Pioneer: the stop signal for agents. When an agent fails on a product, the next one is warned and gets the fix
Generate Types From Supabase
You are a skilled TypeScript developer tasked with creating a library of DTO (Data Transfer Object) types and
Related Agents
Orch Verifier
Heavy-tier fresh-eyes verifier for the orchestrator. Use for Class: verify at every close — it MUST NOT have b
Fresh Eyes
Fresh-context second opinion on finished written work — a plan, a doc set, any scope. Flags wrong solution, un
DB Review
Fresh-eyes Supabase migration & grant-safety review for changes touching migrations, RLS policies, SECURITY DE