What it is
An MIT-licensed GitHub repo containing a SKILL.md + reference files that instruct Claude (or other LLMs) to remove common “AI voice” phrases, structures, and rhythms from prose.
Gabriel’s notes
Stop Slop is an open-source “skill” (a SKILL.md file plus supporting reference Markdown) intended to help Claude (or any LLM) spot and remove common AI-writing patterns—predictable phrases, formulaic structures, and overly performative rhythms. The repo’s README also describes multiple ways to use it (Claude Code skill folder, Claude Projects, custom instructions, or via API/system prompt) and includes a simple scoring rubric so you can tell whether the rewrite actually improved the text.
Quick take: This is the rare “anti-AI-tells” resource that’s concrete: it doesn’t just scold you for being vague—it hands you a repeatable checklist and a pile of patterns to hunt down. Used gently, it makes drafts less corporate, less breathy, and less “hello fellow humans.” Used aggressively, it can sand off nuance and safety language.
I saved this under Prompting because it’s basically a reusable instruction pack: you’re not installing software so much as installing a writing policy into your model’s head.
My review snapshot (from my own notes):
- Source: the upstream GitHub repo.
- Reviewed version (pinned): commit
8da1f030185bdfe8471220585162991eaeb970e9. - Reviewed ZIP SHA-256:
eadc17310d39c1b873728e5e2e3f193fcd7cb7f4015c5f1aaab6ca84d823fd58. - What I saw: a small set of plain-text files (a core
SKILL.md, plus references like phrase lists, structures to avoid, and examples). No “app,” no installer, no runtime.
Good fit if you want to:
- Strip throat-clearing openers and “announcing insight” sentences.
- Break the classic LLM habit of tidy, symmetrical, consultant-y paragraph structures.
- Force more direct claims (and fewer vibes disguised as conclusions).
- Make edits fast by delegating the first pass of “cut the fluff” to the model.
- Standardize a house style across a team that’s quietly using LLMs anyway.
Pricing snapshot (auto-enriched):
Free to use. The repository is published under the MIT License, which is generally permissive for personal and commercial use (keep the license/copyright notice). Paid support, hosting, or enterprise options: Unknown / not confirmed.
Work-use / compliance snapshot (auto-enriched):
This is “just text,” but it’s also operational text: skills are designed to shape agent behavior inside Claude environments. Treat third-party skill files like you’d treat third-party code: review what it instructs the model to do, pin versions, and avoid blind updates—especially if you’re using it in regulated/high-stakes contexts. There’s also a growing body of research warning that SKILL.md-style packages can be abused as a semantic supply-chain vector (i.e., the instructions are the payload, not executable code).
Alternatives (auto-enriched):
- Hemingway Editor: better when you want readability feedback (sentence complexity, passive voice highlights) without imposing an “LLM anti-tells” ideology on every sentence.
- ProWritingAid: better when you want broad writing reports and style feedback across long documents; less focused on the specific modern “AI voice” fingerprints this skill targets.
Before you adopt it:
- Pin the commit you reviewed/approved (or vendor it into your own repo) so “free” doesn’t turn into “surprise.”
- Run it in diff mode (or manually review changes) for anything legal/medical/financial/cybersecurity-related—don’t let it delete your uncertainty language, warnings, or citations.
- Decide your exceptions up front (e.g., quotes, code blocks, brand voice, academic hedging) so the skill doesn’t “improve” the truth out of your writing.
Sources
- https://github.com/hardikpandya/stop-slop
- https://support.claude.com/en/articles/12512180-use-skills-in-claude
- https://arxiv.org/abs/2605.11418
- https://hemingwayapp.com/desktop
- https://prowritingaid.com/pricing