What Are the Banned Phrases in Stop Slop and How to Use Them

Stop Slop maintains a comprehensive inventory of over 100 banned phrases in references/phrases.md that identify AI-generated prose, ranging from throat-clearing openers to business jargon, which you can apply by loading the SKILL.md rules into Claude or any LLM.

Stop Slop is a writing skill for AI assistants that strips clichéd, AI-generated patterns from prose. The repository, hosted at hardikpandya/stop-slop, defines specific banned phrases across nine categories to help produce more direct, authentic text.

The Nine Categories of Banned Phrases in Stop Slop

The complete list of banned phrases lives in references/phrases.md and is organized by semantic purpose. Each category targets specific patterns that signal AI-generated writing.

Throat-Clearing Openers

These introductory clauses preface the main point without adding value. According to the source code in references/phrases.md (lines 7–22), phrases like "Here's the thing:" and "Let me be clear:" delay the actual message and should be cut entirely.

Emphasis Crutches

Redundant phrases that attempt to force emphasis but add no semantic content. Lines 27–34 of references/phrases.md list entries such as "Full stop." and "Let that sink in." as prohibited markers of AI prose.

Business Jargon

Buzzwords that can be replaced with plain language. Lines 35–52 document substitutions like replacing "Lean into" with "Accept, embrace" and "Double down" with "Increase focus on." These terms create corporate vagueness rather than clarity.

Adverbs and Softeners

Any adverbial hedge that dilutes certainty. Lines 53–74 of references/phrases.md ban words like "really," "just," "literally," and "honestly," which weaken statements and create hesitancy in the text.

Filler Phrases

Non-essential opening or closing clauses that pad word count. Lines 75–84 identify "At its core," "In today's [context]," and "When it comes to" as empty transitional devices that delay meaning.

Meta-Commentary

Self-referential asides that announce the structure of the text rather than delivering content. Lines 85–98 list banned phrases including "Hint:", "Plot twist:", and "You already know this, but" as intrusive narrative interruptions.

Performative Emphasis

Statements that attempt to force sincerity through assertion. Lines 101–108 flag "I promise" and "They exist, I promise" as artificial sincerity markers that undermine credibility.

Telling Instead of Showing

Direct statements of difficulty or significance without demonstration. Lines 109–116 prohibit phrases like "This is genuinely hard" and "This is what leadership actually looks like" because they substitute abstract claims for concrete evidence.

Vague Declaratives

Generic importance statements that lack concrete detail. Lines 118–126 ban constructions like "The reasons are structural" and "The stakes are high" without supporting specifics.

How to Apply the Banned Phrase List in Your Workflow

The core rules that enforce these removals are defined in SKILL.md (lines 13–15). To apply the banned phrases in practice, follow this integration workflow:

1. Add the skill to your LLM. Copy the entire stop-slop/ folder into your LLM's skill directory, or upload SKILL.md plus the references/ files as part of your prompt context.
2. Reference the list explicitly. Include an instruction in your prompt to "apply the rules in references/phrases.md" to activate the filtering behavior.
3. Run a pre-processing pass. Submit your raw draft to the model with the skill active; the model will automatically strip banned phrases and rewrite sentences according to the structural rules.
4. Iterate with quick checks. Use the "Quick Checks" section from SKILL.md (e.g., "Any 'here's what/this/that' throat-clearing? Cut to the point.") as a manual verification checklist after the model's first pass.

Code Examples for Implementing Stop Slop

Below are three practical methods to invoke the banned-phrase removal against the hardikpandya/stop-slop rules.

Python API Integration

Use the Anthropic client to apply Stop Slop rules programmatically:

import anthropic  # or the appropriate client library

client = anthropic.Anthropic()

prompt = """
You are a writing assistant equipped with the Stop Slop skill.
Please rewrite the following paragraph, removing all banned phrases
from references/phrases.md and applying the core rules.

> Here's the thing: building products is hard. Let me be clear: it's
> not because the technology is complex. It's because people are complex.
"""

response = client.completions.create(
    model="claude-3-5-sonnet-20240620",
    max_tokens=512,
    temperature=0,
    prompt=prompt,
)

print(response.completion)

YAML Skill Configuration

Add Stop Slop as a persistent skill in a Claude project:

skill:
  name: stop-slop
  path: ./stop-slop
  include:
    - SKILL.md
    - references/phrases.md
    - references/structures.md

Direct API Call

Apply the rules via curl without file uploads:

curl https://api.anthropic.com/v1/complete \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-3-5-sonnet-20240620",
    "max_tokens": 512,
    "temperature": 0,
    "prompt": "Apply the Stop Slop rules (see https://github.com/hardikpandya/stop-slop/blob/main/references/phrases.md). Rewrite:\n\n\"The truth is, we really need to lean into the landscape and double down on the deep dive.\"\n"
  }'

Summary

• Stop Slop bans 100+ phrases across nine categories including throat-clearing openers, business jargon, and performative emphasis, defined in references/phrases.md.
• Core enforcement rules reside in SKILL.md (lines 13–15) and define how the model should process and rewrite text containing banned patterns.
• Integration requires loading both SKILL.md and references/phrases.md into your LLM context, then explicitly instructing the model to apply the filters.
• Verification can be done using the Quick Checks from SKILL.md to catch any remaining throat-clearing or filler phrases.

Frequently Asked Questions

What file contains the banned phrases in Stop Slop?

The complete banned phrase list is stored in references/phrases.md. This file organizes over 100 prohibited expressions into nine semantic categories, with each entry mapped to specific line ranges (e.g., throat-clearing openers occupy lines 7–22).

How do I integrate Stop Slop banned phrases into Claude?

Upload the SKILL.md file and the references/ directory to your Claude project, then reference the files in your prompt instructions. According to the repository structure, you can also copy the entire stop-slop/ folder into Claude's skill directory for persistent access across conversations.

Can I use Stop Slop with other LLMs besides Claude?

Yes. While the documentation targets Claude's skill system, you can use the references/phrases.md file with any LLM by including its contents directly in your system prompt or context window. The banned phrase rules are model-agnostic and rely on standard prompt engineering rather than proprietary features.

What is the difference between throat-clearing openers and filler phrases?

Throat-clearing openers (lines 7–22) are introductory clauses that announce the upcoming point, such as "Here's the thing:" or "Let me be clear:". Filler phrases (lines 75–84) are broader transitional crutches that pad sentences, including "At its core" or "When it comes to." Both delay meaning, but throat-clearers specifically pretend to introduce significance while filler phrases merely occupy space.

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