Slopbeth

Slopbeth

Slopbeth removes AI slop without changing what the draft can honestly claim.

It does not "humanize" text by adding quirks. It cuts unsupported claims, preserves facts, protects voice, and leaves already-good writing alone.

proof

Slopbeth is ranked first in the public anti-slop comparison because it ships rules and evidence together: an 88-case v2 output corpus, 264 judge rows, span annotations, false-positive rows, cadence checks, competitor-output panels, and a 25-case real competitor-agent panel from a remote test host.

Current v2 result:

• Slopbeth wins 23 of 25 real competitor-agent cases.
• Slopbeth preserves required facts with zero missing required facts in the panel.
• Slopbeth records zero forbidden-output hits and zero hard signatures in the panel.
• The release gate requires the v2 corpus, false-positive tracker, span annotations, cadence checks, and competitor-agent panel to pass.

Ranked matrix:

Rank	Repo	Domain	Score	Strongest evidence	Limit
1	Slopbeth 1.3.6	writing	99	88-case v2 output corpus, 264 judge rows, span annotations, false-positive tracker, cadence gate, competitor-output panel, 25-case real competitor-agent panel, score snapshots, installer verification	English-first; detector panel remains weak evidence
2	blader/humanizer	writing	84	broad pattern catalog and false-positive guidance	limited public benchmark evidence
3	d-wwei/great-writer	writing modes	78	mode-specific writing lanes	limited fixture evidence
4	willmather95/human-copy	writing	74	explicit eval checklist	checklist-only; no full release corpus found
5	stephenturner/skill-deslop	scientific prose	72	compact scientific-writing focus and references	no runnable benchmark found
6	sirambrosio/humanink	writing	70	issue-backed false-positive tracker and modal-stacking pattern	pattern scoring can overflag human text
7	hardikpandya/stop-slop	writing	68	compact phrase and structure catalogs, active issue/PR stream	weaker benchmark and fact-preservation evidence
8	jalaalrd/anti-ai-slop-writing	writing	65	compact cross-agent skill and banned-word list	detector claims need stronger caveats

Read the v2 evidence in BENCHMARKS.md.

why it works

Most anti-slop passes chase surface tells. They ban a few phrases, add contractions, vary sentence length, or rough up the prose so it looks less machine-made.

Slopbeth uses a stricter standard:

• keep every sourced fact
• cut claims that have no evidence
• preserve uncertainty, dates, numbers, obligations, and voice
• reject detector tricks that make the writing less true
• leave already-good human text alone
• make the prose dense enough that a summary loses real ideas

The name is a play on Macbeth. Shakespeare's best lines carry pressure and consequence. Slopbeth applies that standard in a narrow way: every sentence should earn its place.

install options

Install into supported agent skill directories:

npx github:ehmo/slopbeth install

That installs Slopbeth for Codex, Claude Code, Hermes, OpenClaw, OpenCode, and Pi. The installer also accepts the installnpx alias for older copied commands:

npx github:ehmo/slopbeth installnpx

Install as Claude Code and Codex plugins:

npx github:ehmo/slopbeth install-plugin

Install only one plugin target:

npx github:ehmo/slopbeth install-plugin claude
npx github:ehmo/slopbeth install-plugin codex

The Claude plugin is installed as a skills-directory plugin. The Codex plugin is installed under ~/.codex/plugins/slopbeth and added to ~/.agents/plugins/marketplace.json.

Marketplace install options:

Claude Code: /plugin marketplace add ehmo/slopbeth
Claude Code: /plugin install slopbeth@slopbeth

Codex CLI: codex plugin marketplace add ehmo/slopbeth
Codex CLI: open /plugins and install Slopbeth from the Slopbeth marketplace

Install somewhere else:

npx github:ehmo/slopbeth install /path/to/skills/slopbeth

If you use the external Skills CLI, request all agents explicitly:

npx skills@latest add ehmo/slopbeth --all --copy

Without --all, skills may install only for the detected current agent.

Direct install:

git clone [email protected]:ehmo/slopbeth.git
cd slopbeth
node bin/slopbeth.js install

Use it:

Use $slopbeth to rewrite this. Preserve facts, dates, numbers, uncertainty, and my voice.

Review with it:

Use $slopbeth to mark unsupported claims, bland-clean sentences, promise changes, and places where the draft sounds like AI.

example: marketing slop

Source:

NovaFlow empowers teams to unlock seamless growth through waitlist page and three-step setup.
The source only confirms waitlist page and three-step setup.

Common cleanup: