SlopOrNot: AI humanizer and on-device AI detector for agents

SlopOrNot is a plugin bundle for Claude, Codex, Hermes Agent, OpenClaw, OpenCode, Cursor, Gemini CLI, and other AI agents.

It ships two skills:

• agentic-humanizer: an AI humanizer that rewrites AI-generated text with a full 5-pass workflow, saved preferences, and optional voice matching.
• slop-check: a one-shot on-device AI detector for AI text detection, AI image detection, readability, Text Cleanup, raw OmniAID scores when explicitly requested, and Pro status.

Agentic Humanizer does not need Slop or Not for its core functionality. Without Slop or Not, it still runs the full rewrite workflow and can match a writing sample. Slop or Not Pro adds the measured on-device AI detector loop: AI score, Flesch-Kincaid readability, Text Cleanup before and after humanization, and a cleanup summary in the final output.

What Agentic Humanizer Does

Agentic Humanizer is not a one-shot paraphraser. Paste a draft, answer a few rewrite preferences, and the skill works through five targeted passes:

1. Pattern surgery against common AI-writing tells.
2. Dialect and tone alignment.
3. Reading-level adjustment.
4. Cleanup-aware targeted editing.
5. Final structural rewrite when earlier passes are not enough.

Voice matching is part of the core skill. Add a writing sample at ~/.agentic-humanizer/voice.txt or pass voice=/path/to/file.txt, and the skill extracts a stylometric fingerprint for rhythm, register, contractions, sentence shape, and concrete phrasing. Voice matching works with or without Slop or Not Pro.

How It Runs

Without Slop or Not

This path runs anywhere the skill can run. It does not require Slop or Not, the slop CLI, or MCP.

• Uses inline settings, a saved profile, or the interview.
• Uses optional voice matching from your writing sample.
• Runs all five rewrite strategies.
• Returns loop history with score and grade shown as n/a.
• Does not claim detector convergence or cleanup stats.

Slop or Not Pro

This runs when Slop or Not Pro is reachable through MCP or the slop CLI on Mac.

• Runs Text Cleanup on the source before the baseline.
• Scores each pass with Slop or Not's local AI text detector.
• Checks Flesch-Kincaid reading grade against your target.
• Runs Text Cleanup again on the selected final text.
• Shows a Text Cleanup summary with hidden-character, punctuation, homoglyph, and dialect-substitution counts.

The only capability Slop or Not adds to agentic-humanizer is local measured feedback and cleanup instrumentation. The humanization workflow itself remains available without it.

Slop Check

Use slop-check when you want local analysis without a rewrite:

/slop-check is this AI? <paste text>
/slop-check what reading grade is draft.md
/slop-check is this image AI? ~/Desktop/art.png
/slop-check clean the invisible characters out of this: <paste text>

slop-check needs Slop or Not Pro because all of its work is on-device AI detector, readability, image, cleanup, or status tooling.

Install

Recommended: plugins

Codex:

codex plugin marketplace add numen-tech/slopornot

Then run codex, open /plugins, switch to the slopornot marketplace, and choose Install plugin.

Claude Code:

/plugin marketplace add numen-tech/slopornot
/plugin install slopornot@slopornot

For non-interactive Claude setup after adding the marketplace:

claude plugin install slopornot@slopornot

Claude Code namespaces plugin skills by plugin name, so use:

/slopornot:agentic-humanizer
/slopornot:slop-check

Direct skill installs and other harnesses use:

/agentic-humanizer
/slop-check

Fallback: direct skill install

Use this path for clients that do not support plugins yet.

For Claude Code, Cursor, or Windsurf:

npx skills add numen-tech/slopornot

For Codex CLI, Gemini CLI, or OpenCode, clone the repo once and copy both self-contained skill directories into the harness skill directory:

git clone https://github.com/numen-tech/slopornot /tmp/slopornot

# Codex CLI
rm -rf ~/.codex/skills/agentic-humanizer ~/.codex/skills/slop-check && \
  mkdir -p ~/.codex/skills && \
  cp -R /tmp/slopornot/skills/agentic-humanizer ~/.codex/skills/ && \
  cp -R /tmp/slopornot/skills/slop-check ~/.codex/skills/

# Gemini CLI
rm -rf ~/.gemini/skills/agentic-humanizer ~/.gemini/skills/slop-check && \
  mkdir -p ~/.gemini/skills && \
  cp -R /tmp/slopornot/skills/agentic-humanizer ~/.gemini/skills/ && \
  cp -R /tmp/slopornot/skills/slop-check ~/.gemini/skills/

# OpenCode
rm -rf ~/.config/opencode/skills/agentic-humanizer ~/.config/opencode/skills/slop-check && \
  mkdir -p ~/.config/opencode/skills && \
  cp -R /tmp/slopornot/skills/agentic-humanizer ~/.config/opencode/skills/ && \
  cp -R /tmp/slopornot/skills/slop-check ~/.config/opencode/skills/

After copying, restart your harness so the skills are discovered.

Claude Desktop

Claude Desktop has no plugin marketplace, so it gets a dedicated bundle. Build the zip from a clone:

git clone https://github.com/numen-tech/slopornot /tmp/slopornot
make -C /tmp/slopornot/claude-skills

This writes claude-skills/agentic-humanizer-desktop.zip. In Claude Desktop, open Settings, then Capabilities, then Skills, choose Upload skill, and select that zip. This build has no harness routing: it runs a built-in interview using Claude Desktop's ask_user_input_v0 prompt, one question at a time, and uses a Slop or Not MCP connector when one is attached.

Optional: Set Up Slop or Not Pro

This section is optional for Agentic Humanizer's core rewrite and voice matching features. Set it up if you want Slop or Not Pro scoring, readability, Text Cleanup, cleanup stats, or the slop-check skill.

1. Install Slop or Not for Mac: https://slopornot.ai/download

2. Open the app and unlock Pro from Settings, then Subscription.

3. Open Settings, then Command Line in Slop or Not for the current CLI setup command. You can also call the bundled binary directly:

   "/Applications/Slop Or Not.app/Contents/MacOS/slop" status --json

4. Optionally register slop mcp with your AI client. See skills/agentic-humanizer/references/slop-mcp-setup.md for per-harness configuration snippets.

Verify:

"/Applications/Slop Or Not.app/Contents/MacOS/slop" status --json

Recent builds print {"pro": true, ...} when Pro is active. Older builds use the legacy "premium": true field. The skill still verifies Pro with a real detector call before using Slop or Not Pro.

Usage

/agentic-humanizer
[paste your AI-generated text here]

Claude Code plugin installs use:

/slopornot:agentic-humanizer
[paste your AI-generated text here]

The skill asks for dialect, reading level, tone, and length preference unless you pass inline overrides or have a saved profile. It may also ask whether to use a writing sample for voice matching.

Output without Slop or Not shows the full workflow without detector claims:

## Humanized text
<the rewritten text>

## Loop history
| Iter | AI score | Grade | Strategy |
|---|---:|---:|---|
| 1 | n/a | n/a | pattern surgery |
| 2 | n/a | n/a | dialect + tone |
| 3 | n/a | n/a | grade gap |
| 4 | n/a | n/a | clean + targeted |
| 5 | n/a | n/a | emergency surgery |

> _Ran without Slop or Not Pro. Add Slop or Not Pro for on-device AI detector scoring, readability checks, Text Cleanup, and cleanup stats: <https://slopornot.ai/download>_

## Highest-impact edits
- ...

Slop or Not Pro output adds on-device AI detector scores and cleanup stats:

## Humanized text
<the rewritten text>

## Loop history
| Iter | AI score | Grade | Strategy |
|---|---:|---:|---|
| 0 | 92% | 11.4 | baseline |
| 1 | 71% | 10.8 | pattern surgery |
| 2 | 48% | 10.4 | dialect + tone |
| 3 | 27% | 9.7 | grade gap |
Converged at iter 3 (<=40% AI, grade target 9-11).

## Text Cleanup summary
| Stage | Invisibles | Punctuation | Homoglyphs | Dialect substitutions |
|---|---:|---:|---:|---:|
| Source cleanup | 1 | 2 | 0 | 0 |
| Final cleanup | 0 | 1 | 0 | 0 |

## Highest-impact edits
- ...

The final output does not expose MCP or CLI backend labels.

Inline Overrides

/agentic-humanizer dialect=us grade=8 tone=casual length=±10 threshold=20 max=7 [paste]

Available flags:

Flag	Effect
dialect=us or dialect=uk	Set the English variant.
grade=N	Set the target Flesch-Kincaid grade.
tone=casual, tone=professional, or tone=academic	Set the rewrite tone.
length=±10, length=exp, or length=trim	Keep length close, allow expansion, or allow trimming.
threshold=N	Override the Slop or Not Pro AI-score target.
max=N	Override the Slop or Not Pro measured-iteration cap.
voice=/path/to/file.txt	Use this voice sample for this call only.
voice=off or voice-skip	Skip voice matching for this call.
skip-interview	Use the saved profile if present, otherwise use defaults.

Saved Preferences And Voice

Agentic Humanizer stores optional profile and voice files under:

~/.agentic-humanizer/

Manage them with:

/agentic-humanizer show profile
/agentic-humanizer reset
/agentic-humanizer set dialect=uk grade=10 tone=casual length=±10
/agentic-humanizer show voice
/agentic-humanizer reset voice
/agentic-humanizer set voice=/path/to/file.txt

The voice sample and fingerprint are local files. Fingerprint extraction runs through your current AI assistant, so privacy follows that assistant's local or cloud setup.

What Slop or Not Adds

Slop or Not Pro makes the humanizer more measurable:

• Local AI score per iteration.
• Local Flesch-Kincaid reading grade per iteration.
• Text Cleanup before and after humanization.
• Cleanup stats that show hidden characters, punctuation artifacts, homoglyphs, and dialect substitutions.
• Pro-gated local tools for slop-check.

Detection runs on-device on Apple silicon. Rewriting still runs in your AI assistant, which may be cloud or local depending on the assistant you use.

Credits And License

The 29-pattern rewrite playbook is from blader/humanizer. It is used under the MIT License.

The supplemental AI-tell checks are SlopOrNot-authored and inspired by Wikipedia:Signs of AI writing. They are a broad field-guide supplement, not a replacement for the licensed 29-pattern playbook.

SlopOrNot is licensed under the MIT License.

Slop or Not is a separate Mac app from Numen Technologies. See https://slopornot.ai.

Contributing

See CONTRIBUTING.md. New harness routing files are welcome. PRs that change the 29-pattern catalogue should sync from the licensed source material rather than diverging.