Back to directory

claude-db

A Claude Code plugin that acts as a database expert: it designs schemas, audits them on separate design-and-integrity and performance-and-scale scores, and can plan safe migrations.

claude-db

claude-db is a Claude Code plugin (also usable cross-agent via Vercel Skills) that reviews database schemas the way a senior DBA would. It reads a project's schema files, ORM definitions such as schema.prisma, or a live catalog, and reasons across modeling, integrity, indexing, concurrency, and migration safety for relational, document, key-value, wide-column, vector, time-series, and graph paradigms.

It scores a schema on two separate axes, Design & Integrity and Performance & Scale, deliberately never blending them, since a schema can be well modeled yet fail under load, or fast yet risk data corruption. Every finding includes observed evidence and a runnable verification command; magnitudes are reported as banded (high, medium, low) rather than fabricated numbers, and a check that needs a live database reports needs_api instead of a silent pass.

It works fully offline by default. Read-only commands include audit, explain, score, next, and checklist; fix previews diffs before writing, and destructive migrations require retyping the object name.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.