SlopScore for Code

Catch AI slop before it merges.

A GitHub Action and CLI that reads the lines a pull request adds and flags what coding agents leave behind: // ... existing code ... placeholders, imports of packages you don't have, empty catch blocks, helpers nobody calls, pasted-twice blocks and TODO stubs. Each PR gets a score, line annotations and a pass/fail check.

Free. Public PRs only. The CLI and Action do the same for private repos inside your own CI.
Or paste a diff (scored in your browser)

What it catches

Placeholder code

// ... rest of the implementation where real code used to be.

Hallucinated imports

Packages missing from package.json or requirements. The door to slopsquatting.

Swallowed errors

Empty catch blocks, catch-and-log-a-string, except: pass.

Unused helpers

New functions nothing in the repo calls.

Duplicated blocks

The same code written twice in one PR.

TODO stubs

throw new Error("Not implemented") in code that ships.

Narration comments

// Loop through the items above every line.

Hallucinated APIs

Math.clamp, fs.promises.exists, os.path.exist.

Giant generated files

900 new lines in one file nobody can review.

Plus debug leftovers, as any / @ts-ignore, commented-out and unreachable code, tautological tests, placeholder keys, generic names and naming drift. All patterns.

Why a gate, not another reviewer

Deterministic

Fixed rules, no model. Same PR, same result, every time, so it can be a required check.

Private

Runs in your GitHub runner. No code leaves your CI. Nothing to approve with security.

Cheap and fast

Zero dependencies, typically under a second. Fails sloppy PRs before a human or a $30/seat AI reviewer spends time on them.

Install in two minutes

# .github/workflows/slopscore-code.yml
name: SlopScore for Code
on: pull_request
permissions:
  contents: read
  pull-requests: write   # only for comment: true
jobs:
  slop:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: kburrus64-max/slopscore-code@v1
        with:
          max: "40"
          comment: "true"

Locally: npx github:kburrus64-max/slopscore-code scores your branch against main. --staged works as a pre-commit hook. Configure with .slopscore-code.json. The open-source release is being prepared.

Pricing

Flat per organization, not per seat. The checks themselves are free forever.

Open source

$0
  • CLI + GitHub Action
  • All rules, unlimited repos and PRs
  • Annotations, job summary, PR comment
  • MIT licensed

Team

$49/mo per org
  • Everything in open source
  • Dashboard: slop trend per repo, author and agent
  • Org-wide rules and thresholds
  • Weekly report to Slack or email
  • Up to 50 developers
Early access soon

Agency

$149/mo
  • Everything in Team
  • Unlimited client organizations
  • White-label PR reports for clients
  • "Clean code" badge for delivered repos
Early access soon
For a 10-developer teamPer monthType
CodeRabbit Essentials ($24-30/dev)$240-300AI reviewer
Greptile Pro ($30/dev + $1/review over 50)$300+AI reviewer
SlopScore for Code Team$49Deterministic gate (use alongside a reviewer, or alone)

Competitor prices from their public pricing pages, October 2026. Full comparison.

FAQ

Does it detect whether AI wrote the code?
No. It flags patterns that are common in agent output and bad in any code. A person who writes catch (e) {} gets flagged too.

Which languages?
JavaScript, TypeScript and Python get every check, including import and unused-helper checks. Go, Rust, Java, C#, C/C++, Ruby, PHP, Swift and Kotlin get the comment, stub, dead-code and duplication checks.

Will it be noisy?
It only looks at added lines, skips lockfiles, generated folders, docstrings and tests where it makes sense, and info-level findings never fail a build. One-line suppression: slopscore-ignore.

Is my code uploaded?
Not by the Action or CLI. The PR box on this page fetches a public PR's diff from GitHub to score it and doesn't store it.