/zero-slop

Less slop, more pop in all your writing.

Free, open-source agent skill to find and fix AI slop.

3.7k+ npm downloads
Paper stack from the writing demonstration. Full originals, tracked changes, and rewrites follow in the transcript.

Demo transcript. Five illustrative examples, each shown as an original draft, visible changes, and a rewrite.

LinkedIn. Original: Kairoset now exports up to 50,000 records in one batch, representing a significant milestone in our journey. It is important to note that the previous limit was 5,000 records per batch, underscoring our commitment to continuous improvement. Writing score: 77. Lower is better.

LinkedIn. Changes: the film strikes through removed words and underlines additions. Original: Kairoset now exports up to 50,000 records in one batch, representing a significant milestone in our journey. It is important to note that the previous limit was 5,000 records per batch, underscoring our commitment to continuous improvement. Rewrite: Kairoset now exports up to 50,000 records in one batch. The previous limit was 5,000 records per batch.

LinkedIn. Rewrite: Kairoset now exports up to 50,000 records in one batch. The previous limit was 5,000 records per batch. Writing score: 15.2. Lower is better.

Document. Original: As we move forward, actionable next steps include Maya taking ownership of the pricing review. It is important to note that the team will review her proposal on Friday before making a decision. Writing score: 92.7. Lower is better.

Document. Changes: the film strikes through removed words and underlines additions. Original: As we move forward, actionable next steps include Maya taking ownership of the pricing review. It is important to note that the team will review her proposal on Friday before making a decision. Rewrite: Maya owns the pricing review. The team will review her proposal on Friday before it makes a decision.

Document. Rewrite: Maya owns the pricing review. The team will review her proposal on Friday before it makes a decision. Writing score: 9.5. Lower is better.

Blog. Original: In today's fast-paced digital landscape, it is essential to understand the importance of clear onboarding. Our trial followed 24 new hires for two weeks. Twelve needed help finding the expense form, highlighting an opportunity for improvement. Writing score: 77.8. Lower is better.

Blog. Changes: the film strikes through removed words and underlines additions. Original: In today's fast-paced digital landscape, it is essential to understand the importance of clear onboarding. Our trial followed 24 new hires for two weeks. Twelve needed help finding the expense form, highlighting an opportunity for improvement. Rewrite: Twelve of 24 new hires needed help finding the expense form. Our trial followed them for two weeks.

Blog. Rewrite: Twelve of 24 new hires needed help finding the expense form. Our trial followed them for two weeks. Writing score: 13.3. Lower is better.

Research paper. Original: It is worth noting that our observational study of 32 teams found an association between shorter meetings and fewer reported delays. These findings offer valuable insights; the small sample does not establish causation. Writing score: 79.3. Lower is better.

Research paper. Changes: the film strikes through removed words and underlines additions. Original: It is worth noting that our observational study of 32 teams found an association between shorter meetings and fewer reported delays. These findings offer valuable insights; the small sample does not establish causation. Rewrite: In our observational study of 32 teams, shorter meetings were associated with fewer reported delays. The small sample does not establish causation.

Research paper. Rewrite: In our observational study of 32 teams, shorter meetings were associated with fewer reported delays. The small sample does not establish causation. Writing score: 9.5. Lower is better.

Email. Original: I hope this email finds you well. I wanted to reach out and circle back on the contract. Please send your comments by Thursday so we can sign on Monday. Thank you in advance. Writing score: 96.3. Lower is better.

Email. Changes: the film strikes through removed words and underlines additions. Original: I hope this email finds you well. I wanted to reach out and circle back on the contract. Please send your comments by Thursday so we can sign on Monday. Thank you in advance. Rewrite: Please send your comments on the contract by Thursday so we can sign on Monday. Thank you in advance.

Email. Rewrite: Please send your comments on the contract by Thursday so we can sign on Monday. Thank you in advance. Writing score: 9.5. Lower is better.

Edit writing for
Slop score76.3/100Before editingLower is better: fewer flagged writing problems. Average score for our AI Slop test corpus, written with obvious flaws.
Slop score9-21/100Writing by peopleScores from the human-written samples we use to check the tool. Other human writing can score outside this range.
Slop score12.8/100After Zero SlopAverage for our AI Slop test corpus after editing. 100% of the rewrites passed our writing and source-detail checks.
Test rankingPlaced 1stAmong 4 editing skills in this testEach used the same AI model. We ranked them using Zero Slop’s own checks, not an independent reader review.

These are results from a small test, not a guarantee for your writing. In a separate speed test on one Apple silicon Mac, the scoring tool checked a median of 501.8 documents per second. That excludes AI editing time. Read the test methods and results

Slop saves the sender time by passing the effort to the reader.

The recipient has to find the missing details or work out what the message means.

AI slop

Merriam-Webster made slop its 2025 Word of the Year and defines it as digital content of low quality, produced usually in quantity by means of artificial intelligence. In prose, it often appears as stock phrasing, flat rhythm, template structure, and decorative formatting. Researchers found a rise in words associated with LLM use across 15 million scientific abstracts, estimating that at least 13.5% of 2024 abstracts received LLM help.

Workslop

Stanford’s Social Media Lab and BetterUp Labs coined the term in September 2025 for AI-generated work that looks complete but does not give the recipient enough to act.

Patterns readers notice

An analysis of 89,239 Reddit posts found recurring complaints about em dashes, flat rhythm, and formulaic contrasts. The study is exploratory, and none of those features proves AI use. They are prompts to read more closely, especially when several appear together.

How to humanize AI text without changing the facts.

Sources: Merriam-Webster’s Word of the Year (2025), Stanford and BetterUp in Harvard Business Review (2025), Kobak et al., Science Advances (2025), Muñoz-Ortiz et al., Artificial Intelligence Review (2024) on differences in sentence-length patterns, Wikipedia’s editor-maintained catalog of AI writing signs, and the RAID benchmark (Dugan et al., ACL 2024), which found detector performance weakened across attacks, decoding methods, and unfamiliar models. Zero Slop edits the writing instead of guessing who wrote it. The longer guide covers both terms in full.

A quick AI draft can leave more work for the reader.

In the September 2025 survey of 1,150 U.S. desk workers, 40% reported receiving workslop in the prior month. Recipients estimated almost two hours of repair per incident. These are self-reported estimates, not a measure of all AI work.

Read the workslop evidence and practical checks, with the BetterUp and Stanford survey and its limits.

Examples Before and after.

Illustrative edits across social posts, email, release notes, and technical writing.

Sample rewrites
Before100.0/100
launch post

We’re thrilled to announce that our team has leveraged cutting-edge AI to deliver a seamless onboarding experience, cutting setup time by 40%. This milestone underscores our unwavering commitment to innovation and operational excellence.

After9.5/100
launch post

We cut onboarding setup time by 40% using AI.

Run it on your own draft, right now.

Paste a draft. See what changed, what stayed, and why.

Try it on your writingFree, and no account.

Version
Zero Slop v2.12.11
Scoring
in your browser, nothing sent
Rewriting
hosted; Zero Slop does not store your draft or rewrite
Installed
uses your AI assistant; local checks stay local

Zero Slop MCP

Add to your AI app or agent.

Connect once, then ask your MCP client to deslop a draft. One model call returns the edit, before-and-after scores, and any review warnings.

  • One tooldeslop, with a typed result
  • Checked outputsource-detail problems return clear warnings
  • No keypublic Streamable HTTP, no account required
01

Connect your client

no account or key
Connect Codexabout 10 seconds
codex mcp add zero-slop --url https://mcp.zero-slop.ai/mcp
  1. Run the command in your terminal.
  2. Open Codex and enter /mcp to confirm that zero-slop is connected.
  3. Ask: Deslop this draft.
Official setup reference
Any other MCP client

Add https://mcp.zero-slop.ai/mcp as a remote Streamable HTTP server with no authentication.

View Zero Slop on Glama

Drafts are processed in memory and excluded from the demo cache. Aggregate service data never includes the draft or rewrite. See what is recorded.

The web editor and MCP use Zero Slop’s online editing service. Results and speed may differ when you use the skill in Codex, Claude Code, Cowork, ChatGPT Work, or another assistant, or use a different writing skill.

MCP connectedzero-slop / deslop

YouDeslop this release note.

Inputwriting score 92.7

Kairoset 4.0 is now available, and it is important to note that search is now up to 40x faster; in addition, bulk export supports 50,000 records; furthermore, audit logs retain every change for 90 days on every plan. In conclusion, Kairoset 4.0 is now available.

  1. Score3 tells found
  2. Interpretfacts locked
  3. Rewritesource checks applied
  4. Copy + read aloudclean
  5. Verify + fresh eyesstatus returned
Checked rewritewriting score 9.5

Kairoset 4.0 is now available. Search is up to 40x faster. Bulk export supports 50,000 records. Audit logs retain every change for 90 days on every plan.

facts preservedwarnings shown
Example response. Lower scores are better; any unavailable editorial review is reported alongside the text.

Plugins

Find Zero Slop in the Cursor plugin directory.

Zero Slop for Cursor

View the Cursor plugin

CLI

Edit files or piped drafts.

npx --yes zero-slop@2.12.11 deslop draft.md --genre professional
Files and scripts
Read a file or stdin. Add --json for the full result and --require-approved to exit nonzero when review is needed.
Hosted editing
deslop sends your draft to Zero Slop. Requires Node.js 22+ and internet; no Python or API key.
Offline scoring
score draft.md checks writing locally with Python 3. No rewrite or upload.
CLI reference

REST API

The MCP pipeline, over HTTP.

curl --fail-with-body --max-time 75 \ https://mcp.zero-slop.ai/v1/deslop \ -H 'Content-Type: application/json' \ --data '{"text":"Maya owns the pricing review. The team will decide on Friday.","genre":"professional"}'
One JSON request
POST /v1/deslop accepts text, with optional genre and audience. Up to 20,000 Unicode code points per draft.
Check the result
The 14-field response includes the edit, scores and review status. HTTP 200 does not mean every check passed.
Free, shared capacity
No API key or draft storage. Aggregate metrics remain. Call from your server; handle usage limits.

Benchmarked on public samples of AI writing.

We tested our AI Slop test corpus of deliberately flawed drafts on August 26, 2026. Zero Slop’s edits averaged 12.8 on its own writing score; avoid-ai-writing averaged 23.3. All 4 editors used the same model, with the unedited drafts included for comparison. This historical test does not establish which editor people prefer or how today’s releases compare.

Read the benchmark notes
30,000

public samples of writing scored.

We ran Zero Slop over almost 30,000 public samples. The sets answer different questions, so we publish them as checks, not a single accuracy claim.

The public RAID+ collection contains 7,627 texts written by four AI models. We scored all of them at the review threshold of 25. RAID+ records which model wrote each text, but does not rate the writing. The chart shows how the scores vary; it cannot tell us how often Zero Slop is right.
A second writing check.

We also checked the rewrites with AIStoryHub’s public slop checker.

Every workflow used the same drafts.

This checker was not built by Zero Slop. Its result is a useful cross-check, not a verdict on quality or authorship.

Editing skills, compared.

This table compares the features documented in each skill’s saved version, not overall writing quality. We audited Zero Slop against six other skills: de-slop, avoid-ai-writing, no-ai-slop, humanizer, unslop-text, and stop-slop. Supported means an implementation is documented. Partially supported means the instructions guide the assistant through the check. Compare four open-source AI humanizers and their measured results.

Repository auditAudited August 30, 2026

What each skill actually ships

Scroll the table horizontally to compare all seven skills. This historical editing audit excludes reader-review capabilities that were not assessed for these revisions. First Reader's separate review appears below.

What each skill actually ships
CapabilityZero Slopisatimur/de-slopavoid-ai-writingpetergyang/no-ai-slopblader/humanizerJCarterJohnson/unslop-texthardikpandya/stop-slop
Score a draft without rewriting itSupportedSupportedSupportedPartially supportedNot documentedSupportedPartially supported
A slop score you can trace to each phraseSupportedSupportedSupportedNot documentedNot documentedSupportedPartially supported
Machine-readable batch and CI gateSupportedSupportedNot documentedNot documentedNot documentedSupportedNot documented
Rhythm and formatting checksSupportedNot documentedSupportedPartially supportedPartially supportedPartially supportedPartially supported
Checks key details against the originalSupportedPartially supportedSupportedPartially supportedPartially supportedNot documentedNot documented
Compares multiple rewritesSupportedNot documentedNot documentedNot documentedNot documentedNot documentedNot documented
Separate grammar and read-aloud passesSupportedNot documentedPartially supportedPartially supportedPartially supportedNot documentedNot documented
Separate post-verification fresh-eyes gateSupportedNot documentedNot documentedNot documentedNot documentedNot documentedNot documented
Catches repetition across your draftsSupportedNot documentedNot documentedNot documentedNot documentedNot documentedNot documented
Learns from your edits, privatelySupportedNot documentedNot documentedNot documentedNot documentedNot documentedNot documented
Remembers the fixes you keepSupportedNot documentedNot documentedNot documentedNot documentedNot documentedNot documented
Re-checks what it learned; stale rules expireSupportedNot documentedNot documentedNot documentedNot documentedNot documentedNot documented
Sample-based scoring preferencesSupportedNot documentedPartially supportedNot documentedPartially supportedPartially supportedNot documented
A public test suite for its resultsSupportedSupportedSupportedNot documentedNot documentedNot documentedNot documented
Obfuscation-resistant phrase matchingSupportedNot documentedSupportedNot documentedNot documentedNot documentedNot documented
Scores long documents without a hard cutoffSupportedSupportedNot documentedNot documentedNot documentedNot documentedNot documented
SupportedPartially supportedNot documentedNot assessed

We checked each complete repository at the saved version linked below. Supported means the repository includes code, saved files, or a named workflow for the check. Partially supported means a prompt asks the model to perform the check. Not documented means we did not find the feature in that version; it is not proof that the tool can never perform it. Not assessed means we have not checked that feature in that version.

First Reader or Zero Slop?

Different jobs, not a single ranking. Choose the review you need before choosing the tool.

First Reader
Explore where a draft might lose its audience. Two simulated readers react passage by passage; a separate skimmer checks the opening. Notes and attention strips sit beside the draft. It does not rewrite.
Zero Slop
Find generic phrasing, revise it, and compare the result with the source. Local checks track names, numbers, quotes, and links. Read-aloud and fresh-eyes passes review the complete edit.

First Reader is not included in the rewrite replay of our AI Slop test corpus: editing is outside its documented scope. Simulated reactions are hypotheses, not measured human behavior. Source-detail checks are not proof that meaning survived.

Source review: September 10, 2026 (Pacific). First Reader at f56f4feb. Its helpers run locally; model cost and privacy depend on the assistant you use.

A feature list cannot tell you which edit you will prefer. Checking names and numbers does not establish factual truth or unchanged meaning. Zero Slop’s local profiles can stop selected words from your sample being flagged; they do not capture your full voice. The installed skill can compare multiple rewrites; the hosted editor uses at most one model response.

Which revision of each tool was audited
Open the audit data

How to use your slop score.

One phrase does not make a draft bad, and formal does not mean fake. Context decides what should change.

Review the meaning, not just the score

The goal is clearer language without changed claims. Review the final edit in context.

Check the details against your original

Zero Slop compares names, numbers, quotes and links with your original. Keeping those details does not prove a claim is true or its meaning unchanged. Your AI assistant also reviews the context; you decide what to keep.

Your sample guides your assistant

Give your assistant a writing sample and explain your preferences. You can also create a local profile so selected words from your sample are not flagged. That profile does not capture your full writing style.

How Zero Slop edits your writing.

The installed skill guides eight editorial jobs; the free web editor combines the AI work into one model response. The final checks run once more on the exact text you receive.

Loop 1From your draft to the final editEight editorial jobs. On the free web editor, all five AI jobs share one model response; the other three run locally.
01scorerlocal script · no AI02interpreteryour AI · one response03rewriteryour AI · one response04fact gatelocal script05copy deskyour AI · one response06read-aloudyour AI · one response07verifierlocal checks08fresh eyesyour AI · one response

Optional checks can also look for predictable wording and repetition across 3+ related drafts. Neither changes the score.

Loop 2Learn from edits you sharecompare · confirm · remember · retire
  1. Share the version you kept

    When you share its draft and your final version, Zero Slop can compare what changed.

  2. Repeats become rules

    A change must recur in separate drafts and be checked against human-written samples before it guides future edits.

  3. A memory that stays yours

    Approved guidance stays in local files you can inspect or remove. Guidance that is not confirmed again eventually expires.

↻ improves only from edits you choose to share

Outside the loopsRelease reviewpropose · review · test · version · ship
Your private guidance never edits the shared pattern library. Shared changes are reviewed, tested against human writing, and shipped in a numbered release.

The installed skill can learn from edits you share. Hosted editing does not use this private learning loop.

The scoring tool looks at wording, sentence variety, readability, and formatting, and explains what it found. A lower score means fewer flagged writing problems. It does not tell you who wrote the text.

See how the AI slop score works, including its limits, or read the formula and validation notes.

Works with the agent you already use.

Add the skill to an agent that reads SKILL.md files. It uses that assistant’s model and tools; it does not install a separate AI model.

GitHub
> npx skills add manavmishra/ZeroSlop --global
  • Installing Zero Slop
  • Added SKILL.md
  • Connected to your agent

Less slop, more pop in all your writing.

Compatible agents

  • Codex
  • Claude Code
  • Cursor
  • Gemini CLI
  • OpenCode
  • Warp
  • Zed
  • VS Code
  • Any open-source harness
SKILL.md

Instructions for checking, rewriting, and copy-editing your draft

slop 100.0

slop 9.5 · clear

Install once. Use for all your AI-assisted writing.

Use the install prompt to let your agent choose the right method, or follow the setup instructions for your assistant.

Ask your agent to edit your writing with Zero Slop.

Free · open source · MIT

npx skills add manavmishra/ZeroSlop --global

Claude Code · Codex · Cowork · Cursor · Zed · open source harnesses

Claude.ai zip

Common questions.

What the tool does, where it runs, and what it costs.

What is AI slop?

AI slop is writing that sounds finished before the thinking is finished. Common signs include stock phrases, vague claims, repetitive structure, flat rhythm, and formatting that works harder than the ideas. Zero Slop shows you where those problems appear.

Is Zero Slop an AI detector?

No. AI detectors try to identify the author. Zero Slop reviews the writing. It flags passages that deserve attention, edits the draft, and checks names, numbers, quotes, and links against the original.

Is Zero Slop just a prompt?

No. It combines eight editorial jobs: scan, interpret, rewrite, protect source details, copy edit, read aloud, verify, and review with fresh eyes. The free web editor uses one successful model response; the score and source-detail checks run in your browser.

Will it change my facts?

It is designed to catch those changes. Before editing, it records names, numbers, quotes, and links. After editing, it checks them again and warns you about anything added or lost. You still make the final call.

Does the scorer send my writing anywhere?

No. The scorer is a local Python script that uses only the standard library. It works offline and needs no account, server, or network connection. The skill itself runs inside the coding or writing agent you already use.

Where can I use it?

Zero Slop works with agents that support SKILL.md files, including Codex, Claude Code, Cursor, Gemini CLI, OpenCode, Warp, and Zed. The repository also documents setup for ChatGPT and Claude.ai.

Is Zero Slop free?

The skill and local scorer are free under the MIT license. Your assistant may charge for model use and applies its own data policies. Hosted editing is free within shared usage limits; local scoring works offline without an account.

What are the hosted editing limits?

Hosted editing allows at most five model calls per client network per UTC day and two per minute, subject to a global budget. The web editor, MCP, REST API and CLI editing share this allowance. Offline scoring does not consume it. There is no reserved capacity or uptime guarantee.

How does Zero Slop learn from my edits?

In the installed skill, a phrase must recur across 3 separate edits; a single word across 5. New guidance must pass the human-writing safety set before it can affect later edits. You can inspect or remove it. Your AI model is not retrained.

What analytics does this website collect?

The website records aggregate page traffic, performance, chapter views, scroll depth, example choices, and selected controls. The remote MCP separately records aggregate reliability, client family, coarse region, size, score, and latency data. No cookies, persistent identifiers, fingerprinting, or draft text are collected. Neither custom telemetry dataset stores rewrite text, IP addresses, raw user-agent strings, or stable user IDs. Cloudflare still processes standard connection data to deliver and protect the service. Global Privacy Control and Do Not Track disable website interaction events.

Keep only the part you actually want to say.

Zero Slop is free, open source, and installed in one command.

Free · open source · MIT

Run it on every draft.

Edit and check your writing in the AI assistant you already use.

Star it122 ★
Send feedback

Goes straight to the maintainer, nowhere else.

Share Zero Slop.

Zero Slop: Less slop, more pop in all your writing. 100+ stars; 3.7K npm downloads. A paper-stack editing example with zero-slop.ai and github.com/manavmishra/ZeroSlop.

/zero-slop. Less slop, more pop in all your writing.

FacebookLinkedInWhatsAppXBlueskyThreadsEmail