vibecheck

Scan codebases for AI-generated code issues ("AI slop") and generate structured fix prompts.

84% of developers use AI coding tools, but the generated code consistently contains placeholder data, stub functions, empty error handlers, hardcoded secrets, and debug leftovers. vibecheck detects these patterns, scores your codebase with an AI Slop Index (0-100), and generates a fix prompt you can paste into any AI tool to fix its own mistakes.

Install

npm install -g vibecheck

Usage

Scan a project

vibecheck scan

  vibecheck — AI Slop Index

  32  [C+]
  ▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░░░░░░░░░░

  18 issues found  ·  2 critical | 5 high | 8 medium | 3 low
  142 files scanned in 89ms

Generate a fix prompt

vibecheck prompt

Generates a structured markdown prompt with every issue, its location, code snippet, and fix instructions. Paste it into Claude, ChatGPT, Cursor, or any AI tool to fix everything at once.

vibecheck prompt --output fixes.md   # Write to file
vibecheck prompt --clipboard          # Copy to clipboard

JSON output (for CI)

vibecheck scan --json

Returns structured JSON with score, grade, passed, and full issue list. Exit code 1 when score exceeds threshold.

# Fail CI if AI Slop Index exceeds 30
vibecheck scan --json --threshold 30

Initialize config

vibecheck init

Creates a .vibecheck.yaml with default settings.

What it detects

Pattern rules (regex-based)

| Rule | What it catches | Examples | |------|----------------|----------| | placeholder-data | Mock emails, names, URLs, API keys, passwords | [email protected], John Doe, sk-test-xxx | | debug-leftovers | Console statements, TODO/FIXME, debugger, AI artifacts | console.log(...), // TODO: implement, // I've updated the code to... | | hardcoded-values | Localhost URLs, connection strings, secrets, ports | http://localhost:3000, postgres://admin:secret@... | | generic-names | Overly generic variable/function names | const data =, const temp =, function doStuff() |

AST rules (TypeScript Compiler API)

| Rule | What it catches | Examples | |------|----------------|----------| | empty-handlers | Empty catch blocks, console-only error handlers | catch (e) {}, .catch(() => {}) | | stub-functions | Functions returning null/undefined, empty bodies | return null, throw new Error("Not implemented") | | unused-code | Imported identifiers never used in the file | import { unused } from 'mod' |

Scoring

The AI Slop Index (0-100) uses weighted severity scoring:

| Severity | Weight | |----------|--------| | Critical | 10 | | High | 5 | | Medium | 2 | | Low | 1 |

Formula: min(100, round((rawScore / filesScanned) * 5))

Critical issues impose a minimum floor (criticalCount * 5) so they can't be diluted in large projects.

| Grade | Score | |-------|-------| | A+ | 0-5 | | A | 6-8 | | A- | 9-12 | | B+ | 13-16 | | B | 17-22 | | B- | 23-28 | | C+ | 29-34 | | C | 35-40 | | C- | 41-46 | | D+ | 47-52 | | D | 53-58 | | D- | 59-65 | | F | 66-100 |

Configuration

Create a .vibecheck.yaml in your project root (or run vibecheck init):

# Disable a rule
rules:
  generic-names:
    enabled: false

# Override severity
rules:
  debug-leftovers:
    severity: low

# Ignore patterns (added to defaults)
ignore:
  - "generated/**"
  - "vendor/**"

# CI threshold
failThreshold: 30

# Output
output:
  colors: true
  verbose: false

Programmatic API

import { scan, generateFixPrompt, calculateScore } from 'vibecheck';

const result = await scan({ rootDir: './src' });
console.log(result.score, result.grade, result.issues.length);

const prompt = generateFixPrompt(result);

License

MIT