Labsco
numen-tech logo

Slop or Not

from numen-tech

On-device tools to detect AI-generated text and images, score readability, and strip AI artifacts, running locally on Apple silicon.

🔥🔥✓ VerifiedFreeQuick setup
#automation#git#security

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 in English and six other languages, 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 (English only), native-language 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. Variant and tone alignment (language variant and register).
  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.

Languages

Agentic Humanizer supports English, Spanish, German, Italian, Swedish, Danish, and Norwegian Bokmal, plus Norwegian Nynorsk for tells. It detects the source language and confirms it in the interview, loads that language's AI-tell catalogue, and measures readability with the language's native formula (Flesch-Kincaid for English, Wiener Sachtextformel for German, Flesch-Szigriszt for Spanish, Gulpease for Italian, LIX for the Nordic languages). The on-device AI detector is English only, so for other languages the AI score shows as n/a and, with Slop or Not Pro, the loop then converges on the reading-level band instead; the no-Slop core workflow runs all five passes and selects by quality, the same as English. Nynorsk runs with tells but without readability; other languages fall back to language-agnostic structural tells with a clear warning. If the app returns a readability score for an unsupported language, the skill logs it as advisory only and does not use it for convergence.

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 the native reading grade (Flesch-Kincaid for English; Wiener Sachtextformel, LIX, or another formula per language) against your target band.
  • 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.

Humanized text

Language

English (en-US). Readability: Flesch-Kincaid grade.

Loop history

IterAI scoreReadabilityStrategy
1n/an/apattern surgery
2n/an/avariant + tone
3n/an/agrade gap
4n/an/aclean + targeted
5n/an/aemergency 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:

```markdown

## Humanized text

<the rewritten text>

## Language

English (en-US). Readability: Flesch-Kincaid grade.

## Loop history

| Iter | AI score | Readability | Strategy |
|---|---:|---:|---|
| 0 | 92% | 11.4 (College) | baseline |
| 1 | 71% | 10.8 (High school) | pattern surgery |
| 2 | 48% | 10.4 (High school) | variant + tone |
| 3 | 27% | 9.7 (High school) | grade gap |
Converged at iter 3 (<=40% AI, grade target 9 to 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:

FlagEffect
language=<code>Set the target language (for example language=de). Without variant=, uses that language's default variant from the registry, or other:<code> for an unsupported language.
variant=<spec>Set the variant (for example variant=de-AT).
dialect=us or dialect=ukLegacy English alias for variant=en-US or variant=en-GB.
grade=NSet the target Flesch-Kincaid grade (English only).
level=<band>Set the reading-level band for any language (elementary to graduate).
tone=casual, tone=professional, or tone=academicSet the rewrite tone.
length=±10, length=exp, or length=trimKeep length close, allow expansion, or allow trimming.
threshold=NOverride the Slop or Not Pro AI-score target.
max=NOverride the Slop or Not Pro measured-iteration cap.
voice=/path/to/file.txtUse this voice sample for this call only.
voice=off or voice-skipSkip voice matching for this call.
skip-interviewUse the saved profile if present, otherwise defaults in the detected language.

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 language=de variant=de-AT level=high_school 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. A saved profile from an earlier version (without a language field) loads as English and upgrades to the current schema on the next write.

What Slop or Not Adds

Slop or Not Pro makes the humanizer more measurable:

  • Local AI score per iteration.
  • Local reading grade per iteration (Flesch-Kincaid for English; native formula for other languages).
  • 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.