slopwash

A prompt that removes AI-generated writing patterns from any text. Copy it, paste it into your LLM, and get output that reads like a human wrote it.

slopwash.com

What it does

LLMs produce text with recognizable patterns: overused vocabulary ("delve," "tapestry," "navigate the complexities"), structural tics (rule-of-three lists, "not just X but also Y"), fake balance, hedging, sycophancy, em dash overuse, and speculative Hollywood endings. These patterns are statistically detectable and make AI-written text obvious to anyone who reads a lot of it.

Slopwash is a set of rules that tells an LLM to scrub these patterns from its own output. The rules are based on published research, Wikipedia's AI-detection field guide, Mozilla Foundation analysis, and observed patterns across GPT-4o, Claude, Gemini, Llama, and others.

How to use it

Copy and paste

1. Go to slopwash.com
2. Pick one or more personas (optional)
3. Copy the prompt
4. Paste it as a system instruction or at the top of any LLM chat
5. Add the text you want cleaned

MCP server

Connect your editor to the MCP endpoint and let your AI agent fetch the prompt programmatically:

https://slopwash.com/api/mcp

VS Code:

// .vscode/mcp.json
{
  "servers": {
    "slopwash": {
      "type": "http",
      "url": "https://slopwash.com/api/mcp"
    }
  }
}

Cursor:

// .cursor/mcp.json
{
  "mcpServers": {
    "slopwash": {
      "url": "https://slopwash.com/api/mcp"
    }
  }
}

Claude Code:

// .mcp.json
{
  "mcpServers": {
    "slopwash": {
      "type": "url",
      "url": "https://slopwash.com/api/mcp"
    }
  }
}

Tools available:

• get_slopwash_prompt — returns the full prompt. Optional params: personas (array), model (string).
• analyze_text — scores text for AI writing patterns (0–100). No LLM call, pure heuristic analysis. Optional param: model.

Agent instructions (always-on)

Paste the prompt into your editor's instruction file so every response follows the rules:

• VS Code: .github/copilot-instructions.md
• Cursor: .cursorrules
• Claude Code: CLAUDE.md
• Windsurf: .windsurfrules

Slop scanner

slopwash.com/scanner — paste any text and get a heuristic score with per-section breakdown and violation details. Runs entirely in the browser, no LLM calls.

Analyzer

The analyzer engine (lib/analyzer/) is pure TypeScript with zero external dependencies. It checks text against six rule categories:

1. Vocabulary — banned and flagged words/phrases
2. Structure — "not just X but also Y," rule-of-three, rhetorical questions, Hollywood endings, etc.
3. Tone & Voice — sycophancy, hedging, chat residue, sentence length uniformity
4. Formatting — em dash overuse, bold overuse, emoji, title case headings
5. Content depth — weasel phrases, source exaggeration, elegant variation
6. Consistency — false emotional understanding, frictionless adoption language

Model-specific profiles (GPT-4o, Claude, Gemini, Llama) add additional pattern detection.

Project structure

app/
  page.tsx          # main site
  scanner/          # slop scanner
  benchmark/        # model benchmark leaderboard
  api/mcp/          # MCP server endpoint
lib/
  prompt.ts         # the slopwash prompt (CORE_RULES + personas)
  personas.ts       # 13 persona overlays
  analyzer/         # heuristic text analysis engine
    rules/          # 6 rule checker modules
    models/         # model-specific profiles
    scoring.ts      # weighted scoring system
extensions/
  vscode/           # VS Code extension (scaffolding)
  browser/          # Chrome extension (scaffolding)
actions/
  slop-check/       # GitHub Action (scaffolding)
packages/
  analyzer/         # npm package (scaffolding)

Development

npm install
npm run dev

License

MIT