FixMySlop: find the AI slop in a GitHub repo

Paste a repo
Public GitHub repositories. The code is downloaded, read and deleted. Nothing is installed or run.
Get the report
A slop score and grade, findings grouped by file, the exact line and the fix. Download it as PDF or Markdown.
Fix it with your agent
Copy the fix prompt into Claude Code, Cursor or Codex, then scan again to confirm the score went down.

What it checks

Deterministic rules for the mistakes AI coding tools make. Same input, same findings, every time.

UI slop

The generated look: AI gradients and glow in any color, glass cards, blur blobs
Invented stats, ratings and testimonials
Colors and spacing that ignore the design system

Accessibility

Clickable divs and icon buttons without a label
Focus outlines removed with nothing in their place
Images without alt text, links that go nowhere

Code slop

// ... existing code ... where real code was deleted
Placeholder implementations shipped as the real thing
Swallowed errors and empty catch blocks

Backend and tests

Tests that assert nothing
The same function copy-pasted across files
Ignored Go errors, Java handlers that only print the stack trace

Security and packages

Leaked API keys and committed .env secrets
Secrets behind NEXT_PUBLIC_ and VITE_
Imports of packages that don't exist

Docs

README commands that package.json doesn't have
Links to files that aren't in the repo
Template placeholders and pasted chatbot replies

Full depth for React, Next.js, Vue, Svelte, Astro, HTML, CSS, Python and JS/TS. General checks for Go, Java, Kotlin, C#, Swift, PHP, Rust, Ruby and more. Supported languages

Run it where you work

The web scan is a preview. The CLI scans private code on your machine, and your agent can check its own work as it writes.

Python 3.10+. JavaScript projects can use npx fixmyslop with no install step.

pipx install fixmyslop

fixmyslop scan                 # the whole repo
fixmyslop scan --changed       # only what your branch changed
fixmyslop ui                   # the app: browse issues, copy fix prompts
fixmyslop scan -s report.pdf   # or .md, .html, .csv, .json, .sarif
fixmyslop baseline             # existing codebase: report only new slop

Installs a skill, an MCP server and a hook. Every file Claude writes is checked, and only the lines it just wrote are reported back to it before it moves on.

fixmyslop init --claude

Cursor rules and an MCP server with check_slop, fix_safe and explain_rule tools. --agents writes AGENTS.md for Codex, Gemini and Copilot agents.

fixmyslop init --cursor
fixmyslop init --agents

Any client configured with JSON (Cursor, Windsurf, Claude Desktop):

{ "mcpServers": { "fixmyslop": { "command": "npx", "args": ["-y", "fixmyslop", "mcp"] } } }

Findings appear as inline annotations on the pull request. Or run fixmyslop init --github.

# .github/workflows/fixmyslop.yml
on: [pull_request]
permissions: { contents: read, security-events: write }
jobs:
  slop:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }
      - uses: MrSpideyNihal/FixMySlop@v0.6.7
        with: { fail-on: high, changed-only: true }

Checks staged files before each commit.

repos:
  - repo: https://github.com/MrSpideyNihal/FixMySlop
    rev: v0.6.7
    hooks: [{ id: fixmyslop }]

Add the badge

Show your grade in your README. It links to the full report and refreshes every few hours.

MrSpideyNihal/FixMySlop

Scan a repository above to get the badge for it.

How accurate it is

Measured on a pinned set of real repositories, with every number reproducible from the repo.

95%of 80 randomly sampled findings were correct on manual review

All Agrades for the four hand-built control repos, including vercel/commerce

About 2xmore findings per 1,000 lines in AI-built apps than in hand-built ones

0 of 20well-known hand-written READMEs (React, Django, Vite) flagged by the docs rules

Method, raw labels and per-repo results

What it can't catch

FixMySlop finds the tells AI-written code leaves behind. It doesn't know what your code is meant to do, so it won't catch logic that is wrong but looks normal, a test that asserts the wrong thing, or a README that describes features the code doesn't have. A clean scan means none of the known tells, not that the code is correct. Keep reviewing code and running tests.

Privacy

Public GitHub repositories only. Nothing is installed or executed.
The download is deleted as soon as the scan finishes.
Reports are unlisted: anyone with the link can open them, search engines don't index them.
Page views are counted anonymously with Vercel Web Analytics: no cookies, no personal data.
For private code, run the CLI. It works offline and sends nothing anywhere.

Find the AI slop in a GitHub repo

Paste a repo

Get the report

Fix it with your agent