Earlier this month I wrote about my Voice Editor workflow — the 6-pass system for transforming AI output into writing that sounds like me. Part of that workflow was a slop detector that flagged AI tells.
Today I’m releasing the detector as a standalone open source skill: The AntiSlop.
Why Open Source This?
Two reasons.
First, Wikipedia’s Signs of AI Writing page documents 24 patterns that identify machine-generated text. Their WikiProject AI Cleanup is doing the hard work of finding and fixing AI-polluted articles. They need better tools, and I had one sitting in my skill library.
Second, every AI writing tool I’ve seen stops at detection. They flag problems but don’t fix them. Highlighting “delve” in red doesn’t help if you still have to rewrite the sentence yourself.
The AntiSlop does both.
What It Catches
35+ patterns across four categories:
Tier 1 — Remove Immediately:
- delve, game-changer, revolutionary, leverage, unlock potential
- “it’s worth noting,” “in today’s digital landscape”
- moreover, furthermore, cutting-edge
Tier 2 — Suspicious When Repeated:
- here’s the thing, at the end of the day, the bottom line
- paired adjectives (“comprehensive and thorough”)
- template openings (“In this post, we’ll cover…”)
Tier 3 — Watch for Clusters:
- transition word overuse (however, firstly, moving forward)
- corporate buzzwords (robust, seamless, scalable)
Structural Patterns: This is where most detection tools fail. These patterns don’t rely on specific words:
- Staccato fragments: “Short. Punchy. Exhausting.” — AI’s version of bullet points pretending to be prose
- Sentence uniformity: Every sentence 10-15 words, no variation
- Comparator sentences: “This isn’t X. It’s Y.” — rhetorical crutch
- Manufactured personality: Fake developer snark trying to sound human
- Self-promotional framing: Author’s achievements as the headline
The Horoscope Test
The skill opens with a simple question:
“Could anyone have written this, for anyone?”
If yes, it’s slop. Like a horoscope — technically applicable to everyone, resonant with no one.
This catches what phrase matching misses. Content can avoid every Tier 1 word and still fail the horoscope test by being aggressively generic.
Editor Mode
The key difference from other tools: it fixes problems by default.
| Action | What It Does |
|---|---|
| Remove | Tier 1 phrases deleted or replaced |
| Deduplicate | Tier 2 phrases kept once, varied after |
| Combine | Staccato fragments merged into flowing prose |
| Simplify | Comparator sentences become direct statements |
| Vary | Sentence lengths diversified |
Request “audit only” if you just want detection without rewrites.
Example
Before:
Let’s delve into how AI is revolutionizing the landscape. It’s worth noting that these game-changing tools are unlocking potential at scale. The speed is impressive. The quality is enhanced. The adoption is growing.
After:
Here’s how teams are using AI coding tools. In a 2024 Google study, developers completed simple functions 55% faster, but showed no improvement on debugging. The tools handle boilerplate well — config files, test scaffolding, repetitive refactors — but can’t tell when they’re wrong.
Same information. Different voice. Specific instead of generic.
Get It
Installation:
mkdir -p ~/.claude/skills
git clone https://github.com/aplaceforallmystuff/the-antislop.git ~/.claude/skills/the-antislopThen in Claude Code: /antislop [your text]
The Workflow Now
My current editing pipeline:
- Draft — Let Claude write freely
- AntiSlop — Catch and fix AI tells
- Voice Editor — Match my specific voice profile
- AntiSlop — Final pass for anything voice editing introduced
Two tools, one job each. The result sounds like me, not like a committee.
Credits
Research backing this comes from:
- Finnish study on “delve” usage (56,878 essays) — 10.45x increase post-ChatGPT
- Georgia Tech analysis (168.3M articles) — “delve” usage went from 0.31 to 7.9 per 1,000 papers
- Biomedical study — “delve,” “realm,” “underscore” co-usage up 85x in 2023-2024
- Wikipedia’s documentation of AI tells
The patterns aren’t opinion. They’re measured.
Related: Voice Editor skill for matching your specific writing voice after slop removal.