Supabash

supabash

A Claude Code skill that generates a **project-specific** `db-cli.sh` for full Supabase PostgreSQL access — direct `psql`, schema migrations, backups, restores. No MCP. No Supabase CLI install. ~200 lines of bash, tailored to your project's paths and schema.

You:     set up supabase db access
Claude:  [reads your repo structure]
         Detected Next.js monorepo at apps/app/.env.local,
         migrations at apps/app/supabase/migrations. Proceeding.
         [generates db-cli.sh + DB_ACCESS_GUIDE.md, updates CLAUDE.md]

You:     ./db-cli.sh tables

Why a skill, not an MCP?

• Zero runtime overhead — no server process, no memory footprint, no JSON-RPC handshakes. MCP tools take up context and resources every session.
• Auditable — the generated db-cli.sh is ~200 lines of plain bash sitting in your repo. You can read it, modify it, commit it, version it.
• Works without Claude — once generated, the script runs fine in any terminal. Your future self (or a teammate) doesn't need an AI agent to use it.
• Per-project customization — the skill reads your actual env paths, migration dir, and schema, then bakes them into the script. No autodetection cruft in the generated output.

Install

As a Claude Code skill

git clone https://github.com/jeep15cba/supabash ~/.claude/skills/supabash

Then in any Supabase project, ask Claude Code:

"set up supabase db access" / "give me psql access to my database" / "generate a db cli for this project"

Claude will invoke the skill, inspect your project, and generate everything.

Without Claude Code

Read [`SPEC.md`](./SPEC.md) and ask any capable LLM (Claude.ai, ChatGPT, Cursor) to generate `db-cli.sh` for your project. Paste SPEC.md + a description of your env file path and migrations dir — the spec is self-contained.

What gets generated

In your project:

1. db-cli.sh at the repo root — bash script with your paths hardcoded
2. DB_ACCESS_GUIDE.md at the r