A Claude skill for removing AI writing patterns from prose.
When you write, draft, edit, or review text, deslop identifies and eliminates predictable AI tells: formulaic sentence structures, filler phrases, false agency, dramatic fragmentation, vague declaratives, and dozens of other patterns that signal machine-generated writing.
The skill works across any prose context. Examples are weighted toward scientific writing and technical blog posts, but the rules apply to any writing where you want a human voice without the AI veneer. For scientific contexts specifically, the skill accounts for conventions like passive voice in methods sections and domain-specific terminology.
Option 1: Claude Code marketplace (recommended)
Install directly from the GitHub repository using Claude Code's plugin marketplace:
/plugin marketplace add nm-dcs/skill-deslop
/plugin install deslop@skill-deslopOption 2: Download ZIP
Click the green Code button at the top of this repo, then Download ZIP. Extract the ZIP and add the folder to your Claude skills directory.
Option 3: Releases
Go to the Releases page and download the latest .skill file. Add it to your Claude skills in customize/skills on the web, or double-click it if you have Claude Desktop installed.
Option 4: Build it yourself
Build a .skill file from the source code, then add it to your Claude skills as described above.
git clone https://github.com/nm-dcs/skill-deslop.git
cd skill-deslop
zip -r deslop.skill SKILL.md references/Install as described above, then use it like you normally talk to Claude. The skill triggers automatically when you:
- Ask Claude to write prose (blog posts, essays, articles, memos, newsletters) making it sound natural instead of AI-generated
- Ask Claude to "deslop", "de-AI" or "make it sound human"
- Ask Claude to check for "slop" or AI patterns
You can also reference the skill directly:
- "Review this draft using the deslop checklist"
- "Score this text on the deslop rubric"
- "Rewrite this paragraph to pass the deslop quick checks"
The skill includes a 1-10 scoring rubric across five dimensions:
| 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.
skill-deslop/
├── .claude-plugin/
│ └── marketplace.json # Claude Code marketplace catalog
├── plugins/
│ └── deslop/
│ ├── .claude-plugin/
│ │ └── plugin.json # Plugin manifest
│ └── skills/
│ └── deslop/
│ ├── SKILL.md # Core rules, quick checks, scoring rubric
│ └── references/
│ ├── phrases.md
│ ├── structures.md
│ ├── tropes.md
│ └── examples.md
├── SKILL.md # Also at root for .skill zip builds
├── references/ # Also at root for .skill zip builds
├── README.md
└── LICENSE
This skill was built in part by combining and synthesizing material from two open sources:
- AI writing tropes catalog from tropes.fyi by Ossama Hassanein. The
references/tropes.mdfile is adapted from this source, and trope patterns are integrated throughout the other reference files. - stop-slop from github.com/hardikpandya/stop-slop by Hardik Pandya. The phrase lists, structural patterns, before/after examples, scoring rubric, and quick checks draw from this project.
MIT