ItsssssJack/SlopMonster

What it solves

SlopMonster is designed to remove the "AI smell" from written copy—the predictable vocabulary, sentence structures, and rhythms that make AI-generated text feel untrustworthy to human readers. It transforms generic, machine-like marketing copy into specific, human-sounding prose that focuses on evidence and utility rather than abstractions.

How it works

The project implements a "de-sloping" loop consisting of four stages:

1. Score: A Python-based linter (deslop.py) analyzes text against a catalogue of AI patterns. It assigns a score out of 5; any score below 5 causes the command to exit with an error, allowing it to be used as a build gate in CI/CD pipelines.
2. Rewrite: The copy is rewritten to remove AI vocabulary and shapes.
3. Cleanse: A shell script (cleanse.sh) sends the draft to a different model family than the one that wrote the original (e.g., if Claude wrote it, GPT cleans it) to strip remaining AI tells.
4. Rescore: The text is scored again to ensure it reaches a 5/5 before shipping.

Who it’s for

It is for developers, marketers, and writers who use AI to draft landing pages, READMEs, emails, and scripts, and want to ensure the final output does not sound like a machine.

Highlights

• Pattern-Based Scoring: Uses regex and root-matching to catch AI-specific vocabulary (e.g., "delve", "seamless") and sentence shapes (e.g., "not just X, but Y").
• Build Integration: Can be integrated into GitHub Actions to fail builds if the copy doesn't meet the quality threshold.
• Model-Aware Cleansing: Specifically routes text to rival AI model families to prevent a model from "marking its own homework."
• Proof Verification: Flags fabricated social proof (like "10,000+ happy users") to force the use of real, verifiable evidence.
• Agent Integration: Can be installed as a skill for Claude Code or other AI agents via SKILL.md.