Who is stop-slop for?

Writers, marketers and developers who use LLMs to draft text and want it to sound less generic; you only need to know how to add a skill or system prompt to your tool.

What can you build with stop-slop?

Edit a blog post draft so it loses the usual AI phrasing before you publish it. Add it to a Claude Code or Claude Projects setup so every piece of writing follows the same style rules. Score your own writing against the five dimensions to see where it feels flat or padded. Study the phrase and structure lists to learn which habits make prose sound machine-written. Include it in an API system prompt so a writing tool produces cleaner first drafts.

hardikpandya /

stop-slop

A skill file that teaches an LLM to strip predictable AI patterns from its writing

18.3k stars
548 stars since 3 Oct
Updated 7 months ago
Beginner friendly
Agent skills & plugins

Open on GitHub (opens in a new tab)

AI-generated prose has habits: throat-clearing openers, binary contrasts, short dramatic fragments, and a steady sentence rhythm. Readers notice these quickly. This project collects those habits into a skill that a model reads before it writes or edits.

The core is a SKILL.md file with the main rules. Three reference files sit beside it: a list of phrases to remove, a list of structural patterns to avoid, and before/after examples. The reference files load on demand when used through the API.

It also includes a scoring step. The model rates a draft from 1 to 10 on directness, rhythm, trust, authenticity and density, and revises if the total falls below 35 out of 50. It is plain text instructions, so there is no code to run or maintain.

What it does

Banned phrase list
Covers throat-clearing openers, emphasis crutches, business jargon, adverbs, vague declaratives and meta-commentary.
Structural patterns to avoid
Names patterns such as binary contrasts, negative listings, dramatic fragmentation, rhetorical setups, false agency and passive voice.
Sentence-level rules
Bans Wh- sentence starters, em dashes, staccato fragments and lazy extremes, and requires active voice.
Five-part scoring
Rates directness, rhythm, trust, authenticity and density from 1 to 10. A score under 35 out of 50 means revise.
Works in several setups
Can be added as a Claude Code skill, uploaded to Claude Projects, pasted into custom instructions, or put in an API system prompt.

What you could build

Edit a blog post draft so it loses the usual AI phrasing before you publish it.
Add it to a Claude Code or Claude Projects setup so every piece of writing follows the same style rules.
Score your own writing against the five dimensions to see where it feels flat or padded.
Study the phrase and structure lists to learn which habits make prose sound machine-written.
Include it in an API system prompt so a writing tool produces cleaner first drafts.

Get started

Download the repo folder. In Claude Code, add it as a skill. In Claude Projects, upload SKILL.md and the reference files. For other tools, copy the core rules from SKILL.md into custom instructions or your system prompt.

Commands are copied from the project's README. Read it on GitHub for the full setup.