What Is the Banned Vocabulary List in No AI Slop? Categories and Implementation

The banned vocabulary list in No AI Slop comprises four distinct categories—banned words, filler phrases, often-empty adverbs, and inflated claims—that the skill automatically strips from text unless they appear within quotation marks.

The No AI Slop skill (maintained in the petergyang/no-ai-slop repository) defines a dynamic banned vocabulary framework rather than a static word list. This system identifies and removes AI-generated fluff during the "slop-removal" phase, preserving the writer's authentic voice while eliminating unnecessary modifiers and unsubstantiated claims.

The Four Categories of Banned Vocabulary

According to the skill's evaluation logic in skills/no-ai-slop/eval.md, the banned vocabulary list organizes problematic language into four semantic groups. Each category targets specific patterns that contribute to vague, inflated, or robotic prose.

Banned Words

Individual terms considered filler or vague placeholders fall into this category. The source code specifically identifies words such as "basically", "actually", and "really" as candidates for removal. These terms often pad sentences without contributing substantive meaning, adding unnecessary syllables that dilute the writer's intent.

Filler Phrases

Short clauses that add no informational value comprise the filler phrases category. Examples include "you know", "as you can imagine", and "to be honest". These conversational crutches create the illusion of accessibility but typically signal AI-generated text attempting to simulate casual human speech patterns.

Often-Empty Adverbs

Adverbs that weaken statements without adding precision populate this category. The banned vocabulary list flags intensifiers like "very", "quite", and "somewhat" as problematic. According to the implementation in skills/no-ai-slop/eval.md, these modifiers usually indicate imprecise thinking that could be resolved with stronger verb choices or specific data points.

Inflated Claims

Overstated or unsubstantiated assertions define this category, targeting phrases such as "the best ever", "ground-breaking", and "revolutionary". The skill treats these as banned vocabulary because they make absolute claims without evidence, a common pattern in AI-generated marketing copy that lacks factual backing.

The Quoted Text Exception

The banned vocabulary list operates with a critical constraint: terms appearing inside quotation marks remain untouched. As implemented in the evaluation logic, the rule explicitly allows quoted text to serve as examples or dialogue without triggering removal. This exception ensures that the skill does not sanitize legitimate uses of colloquial language or cited speech.

Implementing Banned Vocabulary Detection

Developers can programmatically enforce the banned vocabulary list using regex patterns that mirror the skill's internal logic. The following Python implementation demonstrates how to construct detection patterns for each category while respecting the quoted text exception:

import re

# Define pattern groups matching the banned vocabulary categories

BANNED_WORDS = r'\b(basically|actually|really|very|quite|somewhat)\b'
FILLER_PHRASES = r'\b(you know|as you can imagine|to be honest)\b'
INFLATED_CLAIMS = r'\b(the best ever|ground[-\s]breaking|revolutionary)\b'

# Combine patterns with negative lookahead for quoted exceptions

BANNED_PATTERN = re.compile(
    rf'''
    (?!["'])                 # Ignore if preceded by a quote

    (?:{BANNED_WORDS}|{FILLER_PHRASES}|{INFLATED_CLAIMS})
    ''',
    re.IGNORECASE | re.VERBOSE,
)

def remove_banned(text: str) -> str:
    """Return text with all banned vocabulary removed."""
    return BANNED_PATTERN.sub('', text)

# Example usage

sample = (
    'I was basically the best ever at this, you know, and it was '
    'very impressive. "Actually, it was revolutionary," she said.'
)

cleaned = remove_banned(sample)
print(cleaned)

# Output: "I was  at this,  and it was . "Actually, it was revolutionary," she said."

This implementation skips any term preceded by quotation marks, reflecting the "unless quoted as examples" clause defined in the source repository. The function can integrate into preprocessing pipelines to ensure every draft undergoes slop removal before publication.

Where the Banned Vocabulary Rules Are Defined

The specific terms and categories comprising the banned vocabulary list reside in three key files within the petergyang/no-ai-slop repository:

• skills/no-ai-slop/eval.md: Contains the definitive checklist driving the skill's evaluation logic, including the four-category banned vocabulary framework at line 23.
• skills/no-ai-slop/SKILL.md: Describes the overarching philosophy of preserving voice while removing slop, contextualizing why certain terms make the banned list.
• README.md: Provides repository-wide context regarding the risk of AI "smoothing away" distinctive vocabulary, establishing the rationale for the banned word categories.

Summary

• The banned vocabulary list in No AI Slop organizes restrictions into four categories: banned words, filler phrases, often-empty adverbs, and inflated claims.
• Terms inside quotation marks bypass filtering, allowing cited examples and dialogue to remain intact.
• The evaluation logic resides primarily in skills/no-ai-slop/eval.md, with implementation details available in SKILL.md.
• Developers can replicate the detection system using regex patterns that combine category-specific word lists with negative lookaheads for quoted text.

Frequently Asked Questions

Does No AI Slop use a fixed dictionary of banned words?

No. Instead of a static dictionary, No AI Slop employs categorical filters that target patterns of weak or inflated language. This approach allows the skill to catch variations and context-dependent usage rather than relying on an exhaustive word list that would require constant maintenance.

Can I use words like "basically" or "very" if I'm quoting someone?

Yes. The banned vocabulary logic explicitly exempts text appearing within quotation marks. According to skills/no-ai-slop/eval.md, quoted examples remain untouched even when they contain terms from the banned categories, preserving the integrity of cited speech and dialogue.

Where are the specific banned terms documented?

The definitive banned vocabulary categories are documented in skills/no-ai-slop/eval.md at line 23. The skills/no-ai-slop/SKILL.md file provides additional philosophical context about why these terms qualify as "slop," while README.md offers high-level repository documentation explaining the skill's vocabulary preservation goals.

How does No AI Slop distinguish between acceptable and banned uses of adverbs?

The skill treats adverbs like "very," "quite," and "somewhat" as often-empty rather than universally prohibited. The banned vocabulary implementation targets these terms for removal unless they appear in quoted text, operating on the principle that such intensifiers usually indicate imprecise writing that would benefit from stronger verbs or specific measurements.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}