How to Create Custom Rules Based on Stop Slop Principles: A Complete Guide

You can create custom rules based on Stop Slop principles by editing the declarative Markdown reference files—no code changes required.

Stop Slop is a data-driven skill framework designed to eliminate AI-writing clichés in LLM outputs. According to the hardikpandya/stop-slop repository, the entire system relies on declarative Markdown configuration rather than imperative logic, making it straightforward to create custom rules based on Stop Slop principles for any writing workflow.

Understanding the Stop Slop Architecture

The project organizes functionality into three core components:

• SKILL.md – Declares skill metadata, including the name, description, author, and high-level rules the model must enforce.
• Reference data – Stored in references/phrases.md, references/structures.md, and references/examples.md, these files contain concrete patterns to match and before/after transformation examples.
• Consumer integration – The skill integrates into Claude projects, system prompts, or API calls by injecting the contents of SKILL.md and the reference files.

Because the architecture is entirely data-driven, you can extend functionality without touching core logic.

Step-by-Step Guide to Creating Custom Rules

Follow this seven-step process to add new constraints to your Stop Slop skill.

Step 1: Identify the Pattern Gap

Determine which specific phrase, structural cliché, or rhythmic pattern you want the model to recognize and eliminate from generated text.

Step 2: Add Phrase Rules to references/phrases.md

Append new entries to references/phrases.md using the established tabular format with | Phrase | Problem | columns.

| Phrase                | Problem                     |
|-----------------------|-----------------------------|
| "quickly pivot"       | Business-jargon, vague verb |

Step 3: Add Structural Rules to references/structures.md

Insert new rows under the appropriate section (such as Binary Contrasts or Passive Voice) in references/structures.md.

| Pattern                     | Problem                |
|-----------------------------|------------------------|
| "not X, it's Y"             | Negation-then-assertion |

Step 4: Document Rules in SKILL.md

Add a concise bullet under the Core Rules section in SKILL.md, using markdown link syntax to point to the reference file.

9. **Eliminate negation-then-assertion contrasts.**  
   Replace "not X, it's Y" with a direct statement of Y. See [references/structures.md](references/structures.md).

Step 5: Provide Before/After Examples

Add transformation pairs to references/examples.md to illustrate how the custom rule rewrites offending text.

**Before:** Not only did the team miss the deadline, it also exceeded the budget.  
**After:**  The team missed the deadline and exceeded the budget.

Step 6: Test the Rule

Load the updated skill into Claude or your preferred LLM interface. Run prompts containing the targeted pattern and verify the output respects your new constraint.

Step 7: Version-Control Your Changes

Commit modifications to your repository with descriptive messages. This allows downstream users to track the evolution of your custom rule set.

Essential Files for Rule Development

File	Role
SKILL.md	Core skill metadata and high-level rule enumeration
references/phrases.md	Catalog of banned filler phrases and jargon
references/structures.md	Structural clichés and grammatical patterns to avoid
references/examples.md	Before/after transformation demonstrations
README.md	Integration guide and project overview

Best Practices for Custom Rules

• Keep patterns atomic. Limit each row to one specific phrase or structural pattern to simplify maintenance and debugging.
• Use negative examples sparingly. The skill operates by removing patterns; avoid adding "do-not-use" examples that could confuse the model.
• Maintain table alignment. Align columns with existing tables in the reference files so contributors can scan patterns quickly.

Summary

• Stop Slop uses data-driven Markdown files rather than code for rule definition.
• Create custom rules based on Stop Slop principles by editing references/phrases.md and references/structures.md.
• Always update SKILL.md to document new constraints and link to reference data.
• Test rules by loading the skill into an LLM and verifying output transformations.
• Version control all changes to maintain portability and shareability.

Frequently Asked Questions

Do I need programming skills to create custom Stop Slop rules?

No programming is required. Because the hardikpandya/stop-slop repository implements rules as structured Markdown tables and lists, you only need to edit text files. The skill interprets these declarations directly without requiring code compilation or deployment.

What is the difference between phrase rules and structural rules?

Phrase rules target specific lexical items or jargon strings (e.g., "quickly pivot") and live in references/phrases.md. Structural rules target grammatical patterns or syntactic constructions (e.g., "not X, it's Y") and reside in references/structures.md. Use phrase rules for vocabulary issues and structural rules for sentence-level patterns.

How do I test custom rules before sharing them with my team?

Load the updated skill into a Claude project or your LLM's system prompt interface. Input text that contains the targeted pattern and verify that the output either eliminates the pattern or follows your specified transformation. Iterate on the rule definitions in references/phrases.md or references/structures.md until the behavior matches expectations.

Can I distribute my custom rules as a standalone package?

Yes. Because Stop Slop is self-contained, you can distribute your modified SKILL.md and references/ directory as a standalone skill. Other users can drop these files into their projects or API prompts without installing dependencies, making the custom rules fully portable.

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