eat-spaghetti

Claude Code plugin for detecting and cleaning spaghetti code without breaking dependents.

Scoped analysis with quality score, characterization tests as safety net, and micro-step refactoring verified by loop-test.

Install

npm install -g @mdices/eat-spaghetti

Restart Claude Code after installing. The skill is immediately available.

Usage

In Claude Code, invoke reactively:

/eat:spaghetti <scope>

Or describe naturally: "audita o módulo de carrinho", "limpa esse arquivo".

A post-edit hook can also be configured for proactive mode — see plugins/eat/skills/spaghetti/hooks/post-edit-check.md.

Manual install (GitHub marketplace)

Add to ~/.claude/settings.json:

{
  "extraKnownMarketplaces": {
    "mdices": {
      "source": { "source": "github", "repo": "mdices/eat-spaghetti" }
    }
  },
  "enabledPlugins": {
    "eat@mdices": true
  }
}

What it does

• Phase 0 — discovers scope (files + dependency tree, max 30 files)
• Phase 1 — scores code quality (0–1) and lists findings by severity
• Phase 2 — generates characterization tests to lock current behavior
• Phase 3 — applies one refactoring at a time, runs tests between each step
• Phase 4 — final validation and report

Never changes observable behavior. Never invents new abstractions.