Stop Writing Like a Robot! Use skill-deslop Instead

Your latest paper just got rejected. The reviewer didn't even read past the abstract. "Reads AI-generated," they scribbled. Your grant proposal? Desk-rejected. That blog post you spent hours refining? Crickets. Zero engagement. Here's the brutal truth: your writing has "slop"—those invisible fingerprints that scream "ChatGPT wrote this." And in 2024, readers, editors, and algorithms can smell it from a mile away.

But what if you could keep AI's speed while ditching its robotic signature? What if every sentence you produced felt authentically yours—without sacrificing the productivity boost? Enter skill-deslop, a Claude skill that's secretly becoming the weapon of choice for scientists, technical writers, and content creators who refuse to sound like everyone else. Built by Stephen Turner and open-sourced on GitHub, this tool doesn't just edit your text—it surgically removes the mechanical residue that AI leaves behind. Whether you're drafting a Nature submission or a Substack newsletter, skill-deslop transforms robotic prose into something unmistakably human. And the best part? It installs in under 60 seconds.

Ready to discover why top researchers are quietly abandoning raw AI output? Let's dive in.

What is skill-deslop?

skill-deslop is a Claude skill—a specialized instruction set that enhances Claude's capabilities for a specific task. Created by Stephen Turner, a bioinformatics researcher and open-source contributor, this skill addresses one of the most insidious problems in modern writing: the homogenization of prose by large language models.

The term "slop" itself is telling. It captures the vague, formulaic, personality-drained output that AI generates when left unchecked. You've seen it. The dramatic one-sentence paragraphs. The false urgency. The "In today's rapidly evolving landscape..." openers. The inexplicable love of em-dashes and rhetorical questions. This isn't just annoying—it's professionally damaging. In scientific publishing, grant review, and technical communication, AI-detectable writing triggers immediate skepticism. Worse, it buries your actual ideas under layers of machine-generated filler.

skill-deslop emerged from Turner's recognition that AI tools are here to stay, but their output needs serious refinement. Rather than abandoning AI assistance entirely, he built a corrective layer—a skill that sits between raw generation and final publication. The project synthesizes insights from two pioneering open-source efforts: the AI writing tropes catalog by Ossama Hassanein and the stop-slop project by Hardik Pandya. By combining their research with his own expertise in scientific communication, Turner created something more systematic than a style guide and more actionable than a checklist.

The skill is particularly tuned for scientific and technical writing—contexts where precision matters but authenticity is equally crucial. It understands that passive voice belongs in methods sections, that domain terminology isn't slop, and that clarity trumps cleverness. But its rules apply universally. Any prose that needs to sound like a specific human wrote it can benefit from deslop's intervention.

Key Features That Make skill-deslop Essential

What separates skill-deslop from generic "humanize" prompts or manual editing? Systematic pattern detection combined with contextual intelligence. Here's what you're getting:

• Comprehensive Pattern Library: The skill catalogs dozens of AI writing tells across four reference files—phrases, structures, tropes, and before/after examples. It doesn't just flag problems; it shows you exactly what natural alternatives look like.

• Five-Dimensional Scoring Rubric: Every piece of text gets evaluated on Directness (statements vs. announcements), Rhythm (varied vs. metronmic), Trust (respecting reader intelligence), Authenticity (specific human voice), and Density (cuttable content). The 1-10 scale per dimension gives you an objective 50-point benchmark. Score below 35? Time to revise.

• Automatic Triggering: The skill activates naturally when you ask Claude to write prose, "deslop" text, "de-AI" content, or "make it sound human." No complex syntax to memorize.

• Scientific Writing Awareness: Unlike generic tools that blindly strip passive voice or enforce conversational tone, skill-deslop preserves disciplinary conventions. Methods sections keep their passive constructions. Technical terms remain precise. The skill knows the difference between legitimate formality and artificial bloat.

• Cross-Context Flexibility: Blog posts, essays, articles, memos, newsletters, peer reviews—whatever you're writing, the skill adapts. The examples in its reference files span multiple prose genres.

• Open-Source Extensibility: Licensed under MIT, you can inspect every rule, modify the reference files, or build your own variations. The skill structure is transparent and hackable.

Real-World Use Cases Where skill-deslop Shines

1. Academic Paper Revision

You drafted your introduction with AI assistance. It's coherent, comprehensive—and completely forgettable. Every sentence starts with a transition word. The conclusions are overstated. skill-deslop identifies the formulaic structures and replaces them with prose that sounds like you actually thought through the implications. Your reviewers focus on your science, not your syntax.

2. Grant Proposal Polish

Funding agencies are increasingly skeptical of AI-generated applications. A proposal that screams "ChatGPT" triggers unconscious bias. skill-deslop strips the dramatic fragmentation and false agency—those "We will revolutionize..." claims that no serious researcher actually makes—replacing them with measured, credible statements that respect reviewers' intelligence.

3. Technical Blog Posts

Your engineering blog needs personality to build audience trust. But AI-generated posts all sound the same: overly enthusiastic, weirdly philosophical, stuffed with unnecessary metaphors. skill-deslop restores your authentic voice—the specific expertise and perspective that makes readers subscribe instead of bounce.

4. Peer Review and Editing

When reviewing colleagues' papers or editing team documentation, skill-deslop provides an objective framework. Instead of vague feedback like "this feels off," you can reference specific dimensions of the scoring rubric. "The rhythm is metronomic—vary your sentence lengths." "The density score suggests 20% of this paragraph is cuttable."

5. Newsletter and Memo Writing

Internal communications and subscriber content live or die on authenticity. Readers can unsubscribe, ignore, or worse—mentally filter your messages as "just another AI thing." skill-deslop ensures your updates sound like they came from a specific person with specific expertise, not a generic language model.

Step-by-Step Installation & Setup Guide

Getting skill-deslop running takes under a minute. Choose your preferred method:

Option 1: Download ZIP (Fastest)

Click the green Code button at the top of the GitHub repository, then Download ZIP. Extract the ZIP and add the folder to your Claude skills directory.

Option 2: Releases (Recommended)

Navigate to the Releases page and download the latest .skill file. Add it to your Claude skills at claude.ai/customize/skills on the web, or simply double-click it if you have Claude Desktop installed.

Option 3: Build From Source (Developers)

For those who want to inspect or modify the skill:

# Clone the repository to your local machine
git clone https://github.com/stephenturner/skill-deslop.git

# Navigate into the project directory
cd skill-deslop

# Package the skill files into a .skill archive
# The SKILL.md contains core rules; references/ holds pattern libraries
zip -r deslop.skill SKILL.md references/

After building, add deslop.skill to Claude through the web interface or desktop app.

Verification

Once installed, test with a simple prompt: "Review this sentence for slop: 'In today's rapidly evolving landscape, it is crucial to leverage cutting-edge methodologies.'" Claude should flag multiple issues and suggest natural alternatives.

REAL Code Examples from the Repository

Let's examine the actual skill structure and usage patterns from the skill-deslop repository:

Example 1: Directory Structure

The skill's organization reveals its systematic approach:

deslop/
├── SKILL.md              # Core rules, quick checks, scoring rubric
├── README.md             # Project documentation
└── references/
    ├── phrases.md        # Phrases to remove or replace
    ├── structures.md     # Structural patterns to avoid
    ├── tropes.md         # Full catalog of AI writing tropes
    └── examples.md       # Before/after transformations

What's happening here? This isn't a monolithic prompt—it's a modular knowledge base. The SKILL.md file contains the operational logic: when to trigger, how to score, what priorities to apply. The references/ directory functions as a specialized corpus that Claude consults during execution. This separation of concerns means you can update phrase lists without touching scoring logic, or add domain-specific tropes without breaking core functionality.

The phrases.md file catalogs specific lexical patterns—"In today's world," "It's important to note," "delve into," "leverage." The structures.md file captures syntactic patterns: dramatic one-sentence paragraphs, excessive use of em-dashes, rhetorical question chains. tropes.md (adapted from tropes.fyi) documents higher-level narrative patterns like false modesty, artificial urgency, and vague declaratives. Finally, examples.md provides concrete before/after pairs that train Claude's transformation behavior.

Example 2: Build Command

From the installation section, the source build process:

# Clone the repository from GitHub
git clone https://github.com/stephenturner/skill-deslop.git

# Enter the project directory
cd skill-deslop

# Create the distributable .skill file
# -r flag enables recursive inclusion of directories
# SKILL.md is the primary instruction file
# references/ contains all pattern libraries
zip -r deslop.skill SKILL.md references/

Why this matters: The .skill format is essentially a ZIP archive with specific contents. This build command ensures all reference materials are packaged with the core instructions. The -r (recursive) flag is critical—it pulls in the entire references/ subdirectory tree. Without it, you'd have rules referencing non-existent pattern files. This explicit packaging also means version control: you can tag specific builds, track which reference files changed, and distribute identical skill versions across teams.

Example 3: Scoring Rubric Implementation

The skill's five-dimensional evaluation system:

| Dimension    | Question                                      |
|--------------|-----------------------------------------------|
| Directness   | Statements or announcements?                  |
| Rhythm       | Varied or metronomic?                         |
| Trust        | Respects reader intelligence?                 |
| Authenticity | Sounds like a specific human wrote it?        |
| Density      | Anything cuttable?                            |

Below 35/50: revise.

How this works in practice: When you ask Claude to "score this text on the deslop rubric," it evaluates each dimension from 1-10, then sums for a total. The 35/50 threshold (70%) isn't arbitrary—it represents the point where AI patterns become noticeable to discerning readers. A text scoring 8-9 on Directness gets to the point without throat-clearing. High Rhythm scores show sentence length variation, subordinate clause distribution, and paragraph structural diversity. Trust scores punish condescension and unnecessary explanation. Authenticity is the hardest dimension—it's where the skill distinguishes generic competence from recognizable voice. Density identifies redundancy, which AI generates prolifically when uncertain.

Advanced Usage & Best Practices

Layer skill-deslop with domain expertise. The skill knows general patterns, but you know your field. After deslop processing, add a manual pass for discipline-specific conventions. In bioinformatics, that might mean preserving certain statistical phrasings; in humanities, maintaining theoretical terminology.

Use the rubric iteratively. Don't expect perfect scores on first pass. Run text through, identify your weakest dimension, target revisions, and re-score. Most AI-generated text starts around 22-28/50; two revision cycles typically reach 38-42.

Build team conventions. If your research group or content team adopts skill-deslop, establish shared thresholds. Maybe blog posts need 40+, grant proposals 45+, internal memos 35+. Document your decisions in a team style guide.

Combine with other skills. Claude skills stack. Use skill-deslop for voice, then add a citation-checking skill, then a data-visualization skill. The modular approach beats monolithic prompts.

Contribute back. Found a new AI trope specific to your field? The MIT license lets you fork and extend. Submit pull requests to the main repository—Turner actively maintains the project.

Comparison with Alternatives

The decisive advantage: skill-deslop is the only tool that combines systematic pattern removal with contextual intelligence at zero cost. Generic humanizers swap synonyms blindly. AI detectors tell you there's a problem but don't fix it. Professional editing is excellent but slow and expensive. skill-deslop occupies the sweet spot: fast enough for daily use, rigorous enough for publication, free forever.

FAQ

Q: Will skill-deslop make my writing less clear? A: No—it makes it more clear by removing the vague filler that AI inserts to sound authoritative. The skill preserves and enhances precision.

Q: Does it work for non-English writing? A: The current reference files are English-focused. However, the structural patterns (dramatic fragmentation, false agency) translate across languages. Community forks for other languages are welcome.

Q: Can I use skill-deslop with Claude Free tier? A: Skills require Claude Pro or Team subscriptions. The skill itself is free, but Claude's skill functionality is a paid feature.

Q: Will this help me bypass AI detectors? A: That's the wrong framing. skill-deslop improves your writing quality. Improved writing happens to read as more human because it is more human—more direct, more rhythmic, more authentically voiced.

Q: How often is the skill updated? A: Stephen Turner maintains active development. The repository shows commit history; major updates appear on the Releases page. You can watch the repo for notifications.

Q: Can I modify the scoring rubric thresholds? A: Absolutely. Since it's open source, edit SKILL.md to adjust the 35/50 cutoff or reweight dimensions for your specific needs.

Q: Does skill-deslop work with other LLMs? A: The .skill format is Claude-specific. However, the reference files (phrases.md, structures.md, tropes.md) can be adapted into system prompts for GPT-4, Gemini, or other models.

Conclusion

The AI writing revolution brought incredible productivity gains—and an equally incredible homogenization of voice. Every day, millions of documents emerge sounding identical: the same rhythms, the same filler, the same false urgency. skill-deslop is your escape hatch. It doesn't reject AI assistance; it refines it, stripping away the mechanical residue while preserving the speed and capacity that make these tools valuable.

For scientists, this means papers that get read instead of dismissed. For technical writers, it means building genuine audience trust. For anyone who communicates professionally, it means sounding like yourself—not like a weighted average of internet text.

The installation takes 60 seconds. The improvement is immediate. And at zero cost with full source transparency, there's no reason not to try it.

Stop sounding like every other AI-assisted writer. Start using skill-deslop today. Your readers—and your professional reputation—will thank you.

Found this guide valuable? Star the skill-deslop repository on GitHub and share this article with colleagues who are tired of robotic prose.