How to Integrate Stop Slop with Claude Code: A Step-by-Step Implementation Guide

Stop Slop integrates with Claude Code by loading its SKILL.md file as a system prompt and resolving markdown links to auxiliary reference files that define banned phrases and structural patterns.

The hardikpandya/stop-slop repository provides a markdown-based skill that enforces tighter, human-centric writing standards by eliminating AI-generated clichés. Because Claude Code treats any top-level folder containing a SKILL.md as a reusable skill, integrating Stop Slop requires no source code modifications—only proper placement and loading of the rule files.

Understanding the Stop Slop Skill Structure

Stop Slop is implemented as a skill—a collection of markdown files that define writing-style constraints. The repository structure follows Claude Code's skill conventions, where the main entry point and supporting references are stored in specific file paths:

• SKILL.md – The core skill definition containing high-level rules and the Quick Checks checklist
• references/phrases.md – Lexical "tells" and filler phrases to strip
• references/structures.md – Higher-level structural clichés to avoid
• references/examples.md – Before/after transformations illustrating the rules
• README.md – Overview and licensing information

Claude Code automatically resolves markdown links (e.g., [references/phrases.md]) to load these auxiliary files when processing the skill.

Method 1: Integrate Using the Claude Code Web Interface

The simplest way to integrate Stop Slop is through Claude Code's native skill management system, which detects and loads skill folders automatically.

1. Clone the repository to your local machine:

   git clone https://github.com/hardikpandya/stop-slop.git

2. Add the skill folder to your Claude Code project by clicking "Add skill" and selecting the stop-slop folder. Claude Code recognizes the SKILL.md file at the root and loads it as the system prompt.

3. Verify the rule loading—Claude automatically displays the rule list from SKILL.md and resolves links to references/phrases.md and references/structures.md when generating responses.

4. Invoke the skill by submitting text for rewriting. Claude applies the Stop Slop policy to eliminate banned phrases and structural patterns according to the loaded rules.

Method 2: Programmatic Integration with the Anthropic Python SDK

For automated pipelines or custom applications, load the Stop Slop markdown files directly into the Anthropic API as the system prompt. This method explicitly concatenates the core skill definition with reference files to ensure all rules are available to the model.

from anthropic import Anthropic
import pathlib

# Load the Stop Slop skill files from the repository

repo_root = pathlib.Path("stop-slop")
skill_md = (repo_root / "SKILL.md").read_text()
phrases_md = (repo_root / "references" / "phrases.md").read_text()
structures_md = (repo_root / "references" / "structures.md").read_text()

# Build the system prompt by concatenating core skill + references

system_prompt = "\n\n".join([skill_md, phrases_md, structures_md])

client = Anthropic(api_key="YOUR_ANTHROPIC_API_KEY")

def rewrite_text(text: str) -> str:
    response = client.completions.create(
        model="claude-3-5-sonnet-20240620",
        max_tokens=1024,
        temperature=0,
        system=system_prompt,
        messages=[{"role": "user", "content": text}]
    )
    return response.completion

# Apply the skill to raw text

original_text = """
Here's the thing: we need to double down on the real issue, which is that
the process becomes a bottleneck. The truth is, you can see the problem.
"""
cleaned_text = rewrite_text(original_text)
print(cleaned_text)

The system parameter injects the complete Stop Slop policy, forcing Claude to apply the rewriting rules to every user request in that session.

Referencing Auxiliary Rule Files

When Claude needs concrete examples of banned language, it resolves the relative markdown links defined in SKILL.md to load the reference files:

• references/phrases.md contains specific lexical fillers (e.g., "Here's the thing," "The truth is") that the skill strips from output.
• references/structures.md defines structural clichés such as binary contrasts ("not just X but Y") and negative listings ("It's not about A, it's about B").
• references/examples.md provides concrete before/after pairs that demonstrate the transformation standards.

In the Python SDK example, explicitly concatenating these files ensures the model has immediate access to all banned patterns without needing to resolve file paths during inference.

Validating Output with Quick Checks

After integration, verify that Claude's output respects the Stop Slop rules by running the Quick Checks checklist defined in SKILL.md. You can manually audit outputs against this checklist, or prompt Claude to run the verification automatically:

Check the previous response against the Stop Slop Quick Checks checklist 
in your system prompt and flag any remaining clichés.

This validation step ensures that the integration is functioning correctly and that generated text conforms to the skill's anti-slop standards.

Summary

Integrating Stop Slop with Claude Code requires treating the repository's markdown files as a prompt engineering resource rather than executable code. Key implementation points include:

• Claude Code Web Interface: Add the stop-slop folder as a skill; Claude automatically loads SKILL.md and resolves links to references/phrases.md and references/structures.md.
• Anthropic SDK: Concatenate SKILL.md, phrases.md, and structures.md into the system parameter of client.completions.create.
• File Structure: Maintain the repository's file hierarchy so markdown links resolve correctly to references/examples.md and other auxiliary files.
• Validation: Use the Quick Checks checklist from SKILL.md to verify outputs are free of AI-generated "tells."

Frequently Asked Questions

Do I need programming skills to integrate Stop Slop with Claude Code?

No. The Claude Code Web Interface integration requires no coding—you simply add the stop-slop folder as a skill. The programmatic method using the Anthropic Python SDK requires basic Python knowledge but follows standard API patterns for loading external prompt files.

Can I customize the banned phrases list when integrating?

Yes. Because Stop Slop is pure markdown, you can edit references/phrases.md to add or remove specific lexical items before loading the skill. Claude Code will apply your modified version when you reload the skill or restart the SDK session with the updated file contents.

Which Claude models support the Stop Slop skill integration?

The integration works with any Claude model that accepts system prompts, including Claude 3.5 Sonnet and Claude 3 Opus. The Python SDK example uses claude-3-5-sonnet-20240620, but you can substitute other model identifiers compatible with the Anthropic API.

How does Claude Code resolve the markdown links to reference files?

Claude Code treats relative markdown links (e.g., [references/phrases.md]) in SKILL.md as paths relative to the skill folder root. When loading the skill, Claude automatically reads the linked files from the references/ directory, making the complete rule set available without manual concatenation in the web interface workflow.

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