In the previous two articles, I talked about shuorenhua and Humanizer-zh.
Those were more focused on Chinese writing.
Today’s stop-slop feels different. It’s a Skill for English writing scenarios, designed specifically to catch those obvious AI clichés.
For example, an article starts by clearing its throat:
“Here’s the thing.”
Then it drops a familiar twist in the middle:
“It’s not A, but B.”
And ends with a punchy sentence that sounds like a golden quote.
Each sentence alone isn’t wrong. But together, they create a polished, smooth, yet impersonal article.
stop-slop does something straightforward: it breaks these clichés down into a checklist for AI to check one by one when revising.
It’s not mysterious.
But it works well.
1. Clarifying the Source
The stop-slop repository is here:
https://github.com/hardikpandya/stop-slop
The author is Hardik Pandya.
The repo uses the MIT License.
Its structure is very simple:
stop-slop/
├── SKILL.md
├── references/
│ ├── phrases.md
│ ├── structures.md
│ └── examples.md
├── README.md
└── LICENSE
There are no complex scripts, no command-line tools, and no code calling external models.
It’s purely a rule-based Skill.
SKILL.md contains the general rules.
phrases.md lists phrases to delete.
structures.md lists sentence patterns to avoid.
examples.md shows before-and-after rewrite examples.
So don’t think of it as an “automatic AI-flavor remover” software. It’s more like a pair of editorial glasses.
AI tends to write by following templates. You hand it these glasses, and it knows where to pause.
2. It Watches Words and Patterns
Many tools that remove AI flavor only focus on words.
For example, deleting “empower,” “enhance,” “key,” “important.”
Removing words helps but isn’t enough.
stop-slop cares more about sentence posture. It catches several types of things.
The first type is openings.
For example: “Let me be clear,” “The real issue is,” “It’s worth noting.”
These don’t provide information; they just tell the reader: I’m about to explain.
The second type is emphasis crutches.
Some sentences are weak but lean on phrases like “Remember this,” “This is critical.”
stop-slop’s attitude is simple: a strong sentence doesn’t need you banging a drum beside it.
The third type is business buzzwords.
English has many like “navigate challenges,” “deep dive,” “moving forward.”
Chinese has equivalents: “levers,” “pathways,” “closed loops,” “accumulation,” “empowerment,” “integration,” “efficiency improvement.”
These words can be used. The problem is they often let you avoid specific actions.
The fourth type is sentence templates.
This is one of my favorite parts.
stop-slop doesn’t just say “cut fluff,” it directly calls out structures AI loves:
“Not X, but Y.”
“The problem isn’t X, it’s Y.”
“It looks like X, but actually it’s Y.”
Why are these sentences annoying?
Because they first set up a fake opposite, then present a prepared answer.
Occasional use is fine. But if the whole article is like this, readers feel you’re just showing off your cleverness.
3. How It Works
I understand its rules in four layers.
The first layer looks at phrases.
Which are openings, which are emphasis crutches, which are business buzzwords—mark them first.
The second layer looks at structures.
Are there mechanical reversals, negation parallels, dramatic short sentences, question-and-answer setups?
These issues hide in sentence construction and can’t be fixed by deleting a few words.
The third layer looks at subjects.
stop-slop strongly emphasizes active voice.
Simply put, whoever did the action should be at the front of the sentence.
It also opposes “fake subjects.”
For example, “Data tells us,” “Culture changes,” “Decisions naturally emerge.”
These sound smooth but often hide the real people.
Data doesn’t speak.
Culture doesn’t walk by itself.
Someone reads the data, someone changes behavior, someone makes decisions.
The fourth layer looks at rhythm.
AI loves to write too neatly: three parallel points, a short summary at the end of each paragraph, every judgment turned into a slogan.
stop-slop reminds AI: sentence length should vary, paragraph endings shouldn’t all be punchlines.
This is useful for public accounts too.
Readers don’t want a punchline every three lines.
4. A Concrete Example
For example, this paragraph:
In the context of rapid content creation development, writers need to continuously improve their expressive abilities. The real importance lies not in mastering more tools, but in establishing a stable writing system. Only in this way can one develop their own style through long-term creation.
This paragraph has no factual errors.
But it’s too familiar.
You can almost guess what comes next.
From stop-slop’s perspective, it has at least four problems:
- “In the context of…” is an empty opening.
- “Continuously improve expressive abilities” lacks action.
- “Not… but…” is a mechanical reversal.
- “Only in this way can…” feels like a summary template.
When rewriting, no need to pretend to be profound.
You can write directly:
I’m currently writing a public account and often let AI draft the first version.
The first draft usually looks okay but is too smooth. It lays out background, splits into three points, writes a summary—reading like a standard answer without personality.
So I need a checklist to pick out these cliché sentences first.
This version still has judgment sentences but is more concrete than the original.
It has a scene, actions, and problems.
Readers know what kind of trouble I’m dealing with.
This is the most practical part of stop-slop: it doesn’t turn you into a great writer but helps you catch cliché sentences first.
5. How to Use It
If you want to install it, give this GitHub address to AI:
https://github.com/hardikpandya/stop-slop
Then say:
Help me install the stop-slop Skill from this repository.
After installation, you can use it like this:
Please use stop-slop to check the following passage.
First, list the AI writing clichés detected.
Then provide a more direct, more human-like rewrite.
Do not add new facts or remove the core information.
If you write in English, it can be used directly.
If you write in Chinese, I don’t recommend copying all its rules.
For example, it requires deleting all adverbs. This rule is strict in English but too rough for Chinese. Some Chinese modal particles and adverbs are a normal part of speech.
Also, it opposes all dashes. Occasionally using dashes in Chinese public accounts isn’t necessarily a problem; the problem is using them to fake pauses or fake depth.
So in Chinese contexts, stop-slop is better as a checklist.
You can let AI check once by its rules, then apply Chinese writing habits for a second judgment.
6. What’s Most Worth Learning from This Kind of Skill
When I analyze this Skill, the key is learning how it writes rules.
It doesn’t phrase requirements as vague statements:
Please make the article more natural.
Such prompts are too vague.
AI sees them and can only revise based on feeling.
stop-slop writes like an editorial checklist:
Which phrases to delete.
Which sentence patterns to avoid.
What counts as a fake subject.
What counts as overly neat rhythm.
Which dimensions to score after revision.
This is much more useful than “be more natural” or “less AI flavor.”
Because it breaks a vague feeling into executable actions.
If you want to write your own Skill in the future, you can learn this:
Don’t just tell AI the goal.
Tell it how to judge, how to handle, and how to self-check after handling.
7. My Judgment
stop-slop can’t solve all Chinese AI flavor.
It’s too English and too rigid.
But it’s very suitable as a first filter.
First filter out the most obvious clichés: openings, fake reversals, business buzzwords, golden quote endings, fake subjects, overly neat rhythm.
After filtering, use Chinese intuition to refine slowly.
This is more reliable than just saying “please write more naturally” from the start.
Because “natural” is too broad.
When big words pile up, AI starts performing again.
It’s better to break it down.
Fix one by one.
