Schema To Skill — Data skill for Claude Code

schema-to-skill

Turn your Postgres schema into a skill your AI coding agent actually reads. Stop hallucinated columns.

[](https://github.com/SamRome1/schema-to-skill/actions/workflows/ci.yml) [](https://www.npmjs.com/package/schema-to-skill) [](LICENSE) [](package.json)

`schema-to-skill` connects to your real database, reads the catalog, and writes a `SKILL.md` for Claude Code, a section for `AGENTS.md`, or a Cursor rule. Tables, columns, relationships, enums, functions, triggers, indexes and Row Level Security policies, all in plain English with the exact identifiers. Supabase-aware.

Why

• Agents guess column names. user_id or owner_id? created or created_at? They pick one and move on.
• Agents forget foreign keys and write joins that are almost right.
• Row Level Security blocks silently. The query "works" and returns zero rows, and the agent starts debugging the wrong thing.
• Hand-written schema docs drift the day after you write them.
• This tool generates from the only source that cannot drift: the database itself. Run it again and commit the diff.

Quick start

npx schema-to-skill --db "$DATABASE_URL" --supabase

Until the npm release lands you can run straight from GitHub:

npx github:SamRome1/schema-to-skill --db "$DATABASE_URL" --supabase

That writes `.claude/skills//SKILL.md`. Claude Code picks it up automatically the next time it touches your database code. Commit the folder.

Only reads. Introspection runs inside a `READ ONLY` transaction against `pg_catalog`; nothing is written to your database.

What it generates

An excerpt from a generated skill for the Supabase SaaS fixtu