PlugWiki

Stop-Slop skill for Claude: remove AI tells from your writing

Makes Claude cut AI-sounding phrases and sentence habits from its writing, then score the result and revise it. Plain markdown, no scripts.

Installs
18.6K
on skills.sh
Stars
18.5K
on GitHub
Updated
Mar 2026
last commit

stop-slop makes Claude cut the stock phrases, sentence shapes and rhythms that make writing sound AI-made, then score its own rewrite before handing it back. Ask it to write or edit text and it follows eight rules, runs a 12-point check, and revises anything under 35 out of 50. The rules are absolute (no adverbs, no em dashes, no passive voice), so it can over-trim technical writing. It is plain markdown with no scripts.

Good for

  • Writers who draft with AI and want a final pass before publishing
  • Marketers and founders cleaning up posts, emails and launch copy
  • Claude Code users who want a ban list for prose in every session
  • Editors who want a scored review of a draft without a rewrite

Install stop-slop

Claude Code

  1. 1

    Copy this prompt.

    Install the stop-slop skill in Claude Code. Run this command:

git clone https://github.com/hardikpandya/stop-slop.git ~/.claude/skills/stop-slop

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires git.

The README says to add the folder as a skill, and cloning does that and keeps the reference files. A user reported in July 2026 that the skills CLI route left the references folder out, though the current skills.sh copy includes it. After either route, check that the stop-slop folder has a references folder. To limit it to one project, clone into .claude/skills/stop-slop inside that project instead.

From the project README. Source.

What's in the skill

When it loads

Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells.

Sections in SKILL.md

  1. Core Rules
  2. Quick Checks
  3. Scoring
  4. Examples
  5. License

Key facts

Author
Hardik Pandya
License
MIT
Installs on skills.sh
18,646 (Nov 2, 2026)
GitHub stars
18,524 (Nov 2, 2026)
Files and scripts
7 markdown files, no scripts
Last commit
Mar 17, 2026
Needs
Nothing: no scripts, keys or network
Works in
Claude Code, Claude Projects, Claude API, Cursor (as rules)

What it includes

  • 8 writing rules: cut filler, break formulas, active voice, be specific, vary rhythm

    Eight rules: cut filler, break formulaic structures, use active voice, be specific, put the reader in the room, vary rhythm, trust readers, and cut quotable lines.

    Try: “Write this launch email and follow stop-slop.”

    Core rules (SKILL.md)

  • 12 checks before delivery: adverbs, passive voice, em dashes, Wh- openers

    A 12-item list Claude runs before handing back text. It covers adverbs, passive voice, things doing human verbs, Wh- openers, "not X, it's Y", three same-length sentences, em dashes and vague declaratives.

    Try: “Check this paragraph against the stop-slop quick checks.”

    Quick checks (SKILL.md)

  • Rates directness, rhythm, trust, authenticity, density 1-10; under 35/50 means revise

    Claude scores its text on five points from 1 to 10. If the total is under 35 out of 50, it rewrites and scores again.

    Try: “Score this draft with stop-slop and revise it if it scores under 35.”

    Scoring (SKILL.md)

  • Phrases to cut: throat-clearing openers, emphasis crutches, jargon, adverbs

    Lists of phrases to remove, such as "Here's the thing" and "Let that sink in", plus a table of 11 jargon words with plainer swaps.

    references/phrases.md

  • Sentence shapes to avoid: not X but Y, false agency, passive voice, triads

    Tables of patterns with the problem and the fix. They include binary contrasts, rhetorical setups, false agency such as "the data tells us", and three-item lists.

    references/structures.md

  • 5 before/after rewrites showing the rules applied

    Five rewrites that show the rules at work. Example 4 still uses an em dash, which the rules ban.

    references/examples.md

From the official docs

Permissions

Low risk

stop-slop only changes how Claude writes. If you ask it to rewrite a file, Claude edits that file with its normal permissions.

What it can do

  • Read your data?
  • Create and change things?
  • Delete things?
  • Read-only modeHow to turn it on

Watch out for

  • Rules can cut words that carry meaning

    The ban on adverbs and em dashes is absolute. Read the rewrite before you publish it, especially technical text. github.com

  • No code ships with the skill

    The repo holds SKILL.md, three reference files, README, CHANGELOG and LICENSE. Nothing runs on install. github.com

  • skills.sh audits all pass

    Gen Agent Trust Hub, Socket and Snyk each show Pass (Nov 2, 2026). skills.sh

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details
Turning on read-onlyThe skill is markdown with no scripts, tools or network calls. Any file edits come from the assistant's normal permissions when you ask it to rewrite a file.

What people say

People like the scored second pass and dislike how blunt the rules are.

What people like

  • The 35 out of 50 score gives Claude a stop condition and forces a real second pass.
  • It is small enough to carry anywhere, and one user runs it on everything they publish.
  • Other skill authors reuse its scoring rubric.

Common complaints

  • Blanket rules such as no em dashes and no adverbs cut text that was fine.
  • It over-trims technical docs and product writing.
  • Its own examples break its rules, including an em dash in Example 4.
  • One Hacker News commenter said the README itself reads like AI slop.

In their words

BlogMay 2026
“Without it, "remove AI patterns" is an open-ended instruction a model can nod along to and only half-apply. A numeric floor forces an actual second pass.”
Florian NarrGitHub issueJul 2026
“i've been running a stop-slop pass on basically everything i publish for months. x posts, memos. it started here.”
odinfreeBlogMay 2026
“The rules are also opinionated to the point of rigidity in a few places: "no em dashes at all" and "kill all adverbs" are blunt instruments that will occasionally cut something that was fine.”
Florian NarrGitHub issueApr 2026
“One thing that keeps coming up in practice: the current rules can get too aggressive on technical docs or product writing, where some patterns are fine in moderation but still look like slop in casual prose.”
odinfree
1 more quote
GitHub issueApr 2026
“Several problems but just as a clear issue: the "after" example for the 4th example has an em-dash despite one of the rules being no em-dashes.”
KarmanyaIyer

Examples

  • Clean up a draft written with AI help before publishing

    Run stop-slop on this draft and score it before you hand it back: [paste text]

  • Rewrite a jargon-heavy paragraph

    Edit this paragraph to remove AI tells: In today's fast-paced landscape, we need to lean into discomfort and navigate uncertainty with clarity.

    The skill's own jargon example (references/examples.md) goes from two sentences to "Move faster. Your competition is."

  • Review a post for AI patterns without rewriting it

    Review this LinkedIn post for AI writing patterns and give me the five stop-slop scores.

  • Tidy a chat message before sending it in Microsoft Teams

    Clean up this message for Teams with stop-slop: [paste message]

    One user reports the rewrite can keep hard line breaks mid-paragraph, which show in Teams.

Questions

How do I install stop-slop?

Use the install box above. In Claude Code, cloning the repo into your skills folder is the safest route, because it brings the three reference files along. One user reported in July 2026 that the skills CLI route left the reference files out, and the current skills.sh copy includes them, so check that the folder has a references folder after installing.

Does stop-slop work in Claude.ai?

Yes, through Claude Projects. The author says to upload SKILL.md and the reference files to project knowledge. A zip upload under Customize > Skills also works if you rename the downloaded folder from stop-slop-main to stop-slop first, because Anthropic rejects a folder name that does not match the skill name. The author does not document the zip route.

Does stop-slop work in Cursor, Codex or ChatGPT?

Cursor: the author's launch post says to add the rules to your .cursorrules file or paste them into the system prompt. Codex: the author documents nothing, but Codex's docs load skill folders from a skills directory, so the Codex tab above uses that path. ChatGPT: there is no packaged skill. The README's custom instructions route, copying the core rules from SKILL.md, is the closest fit.

Can I use stop-slop through the API?

Yes, per the README. Put SKILL.md in your system prompt. The reference files load on demand.

Is stop-slop safe?

Yes. It is markdown only, with no scripts, packages or network calls. skills.sh shows three audits, Gen Agent Trust Hub, Socket and Snyk, and all pass (Nov 2, 2026).

stop-slop vs humanizer: which should I use?

Pick stop-slop for a short ban list and a scored final pass. Pick humanizer to rewrite while keeping your facts and matching a sample of your voice. humanizer works from 26 patterns taken from Wikipedia's guide to AI writing, and has about 63,500 GitHub stars and 11,500 skills.sh installs (Nov 2, 2026), against stop-slop's 18,500 and 18,600. Some people run both.

Why does stop-slop not load its reference files?

One user reported that the skills CLI installed SKILL.md without the references folder (issue 46, open since July 2026). The current skills.sh copy of the skill does include all three reference files, so the report and the listing disagree. Clone the repo into your skills folder, or check the installed folder for references.

How does stop-slop score text?

It rates directness, rhythm, trust, authenticity and density from 1 to 10 each. Below 35 out of 50, the assistant revises the text.

Will stop-slop get text past AI detectors?

The author does not claim that. The skill targets habits readers notice, not detector signals. Treat any detector result as a separate test.

Is stop-slop too strict for technical writing?

Often, yes. Users report the rules get too aggressive on technical docs and product writing, and a reviewer notes there is no exception for a word that carries meaning. Pull requests that add a technical mode are open and not merged, so edit the rules or skip the skill for docs.

Information

updated Nov 2, 2026

Repository
hardikpandya/stop-slop

hardikpandya/stop-slopskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 20
  1. hardikpandya/stop-slop · github.com
  2. stop-slop on skills.sh · skills.sh
  3. Bug: installing via npx skills skips references · github.com
  4. Using skills in Claude · support.claude.com
  5. Claude Desktop zip upload frontmatter error · github.com
  6. Build skills (Codex) · learn.chatgpt.com
  7. stop-slop SKILL.md · raw.githubusercontent.com
  8. New Claude skill: Stop Slop · hvpandya.com
  9. phrases.md · raw.githubusercontent.com
  10. structures.md · raw.githubusercontent.com
  11. stop-slop examples.md · raw.githubusercontent.com
  12. Your Examples don't follow the skill rules · github.com
  13. ChatGPT skill zip request · github.com
  14. humanizer SKILL.md · raw.githubusercontent.com
  15. humanizer README · raw.githubusercontent.com
  16. Suggestion: advanced mode with technical-writing exemptions · github.com
  17. Stop Slop: A Skill for Removing AI Writing Tells · codeline.co
  18. your skill became the seed of my writing system · github.com
  19. Stop Slop (HN) · news.ycombinator.com
  20. Rewritten text keeps hard line breaks · github.com

Updated Nov 2, 2026.