ItsssssJack/SlopMonster
Turn AI-written copy into copy a human would ship. Lint for AI tells, cleanse with a rival model, lint again.
SlopMonster – AI‑writing pattern checker
What it does
- A tiny Python‑only tool that scans English text for five kinds of phrasing that often appear in AI‑generated marketing copy (e.g., buzz‑words, “not just X, but Y” constructions, dense punctuation, three‑item promotional lists, and unverified numeric claims).
- It assigns a simple 0‑5 score – one point is deducted for each category that contains at least one match. A perfect 5/5 means none of the known patterns were found.
- The tool can be used directly from the command line on raw strings, Markdown files, or HTML files.
How to use it
# Clone and run the checker on a short snippet
git clone https://github.com/ItsssssJack/SlopMonster
cd SlopMonster
python3 tools/deslop.py --text "It's not just a tool, it's a game‑changing journey."
- Returns exit status 1 for scores < 5, 0 for a perfect score.
- For Markdown it ignores code fences, inline code, images and struck‑through text; for HTML it strips tags and script/style blocks.
Editing workflow
- Run the checker, note the flagged categories.
- Provide the draft and the SKILL.md description to an LLM (Claude, GPT, etc.).
- After the first LLM pass, run the second pass with a different model family using the supplied
prompts/cleanse.txtscript (tools/cleanse.sh). - Re‑run the checker on the cleaned output.
The repository also includes:
tools/cleanse.sh– a wrapper that sends the draft to the chosen CLI model and writes the edited copy to stdout while emitting change notes to stderr.tools/test_deslop.pyandtools/test_cleanse.sh– regression tests.- A GitHub Actions workflow (
.github/workflows/slop.yml) that can fail a CI run when the README or other files score below 5/5.
Limitations
- Works only for English; other languages may slip through.
- It cannot verify factual claims or detect the true author.
- False positives are possible – ordinary phrasing may be flagged, and some poor writing will not be caught.
Who might find it useful
- Teams that want a cheap, zero‑dependency gate to catch obvious AI‑style marketing fluff before publishing.
- Open‑source project maintainers who want CI to block copy that sounds like generic AI‑generated hype.
- Anyone looking for a lightweight, rule‑based “human‑style” checklist to pair with LLM‑based editing.
License – MIT.
Related
- Project
- Project
- Project
- Dispatch
- Project