Plugin
platform-design-skills
Platform design skill pack: 450+ rules for Apple HIG, Material Design 3, and WCAG 2.2 across iOS, iPadOS, macOS, watchOS, visionOS, tvOS, Android, and Web.
8 skills625
@ehmoWriting
Plugin
slopkit
Two anti-slop skills for coding agents, installed by the same means. slopbeth cleans the writing you ship.
Install on Claude Code
> /plugin marketplace add ehmo/slopkit
Stats
110
Stars
7
Forks
Maintained
Maintenance
Python
Language
MIT
License
3mo ago
Last commit
4mo ago
Created
21d ago
Added
What's inside
Skill(2 skills)
/slopbethUse when drafting, editing, reviewing, or benchmarking prose to remove AI-writing tells while preserving meaning, voice, and density. Trigger this skill for requests about AI slop, humanizing AI-assisted writing, detector-facing validation, unsummarizable prose, voice/slopgentslopbeth for the conversation instead of the shipped artifact. Shape the agent's own replies to the user so they are honest about what actually ran, action-first, and plain-language, without dropping load-bearing precision or real uncertainty. Invoke to turn on; it stays until
Badges
README
Slopkit
Two anti-slop skills for coding agents, installed by the same means.
- slopbeth cleans the writing you ship. It rewrites and reviews an artifact, cutting AI tells without changing what the draft can honestly claim.
- slopgent cleans the conversation. It shapes the agent's own replies so they stay honest about what actually ran, action-first, and plain.
They are separate skills. Install both (the default), or remove either one and keep the other. The split is deliberate: point slopgent at a document and you get the clipped formula prose slopbeth exists to remove; ask slopbeth to narrate a build and it has nothing to preserve.
| slopbeth | slopgent | |
|---|---|---|
| Governs | the text you hand over to edit or publish | the agent's own status reports, explanations, and "done" claims |
| Optimizes for | sourced, dense, unsummarizable prose in your voice | honesty first, then structure, then plain language |
| Runtime | Node + Python, with a PowerShell 7 twin | Python (Node for the memory helper) |
| Full docs | skills/slopbeth/README.md | skills/slopgent/README.md |
[!TIP] Not sure which you need? slopbeth edits a document you hand over. slopgent shapes what the agent tells you while it works. Most people want both. The default install adds them as siblings, and deleting either directory leaves the other working.
why
An agent produces two kinds of text, and both drift toward slop in their own way.
- The artifacts it writes for you (the README, the postmortem, the release note) fill with unsupported claims, brand-lesson closers, and sentences that read clean but say nothing. slopbeth cuts that without dropping a single sourced fact.
- The things it says to you ("Fixed the auth bug," "everything should work now") overstate what actually ran and bury the one command you need. slopgent keeps the agent honest about what it verified and leads with the action.
Neither is a "be concise" directive. Concise-only rules cut the filler and the load-bearing caveat with equal enthusiasm. Both skills cut the filler and keep the caveat.
install
Install both skills into supported agent skill directories:
npx github:ehmo/slopkit install
That installs slopbeth and slopgent for whichever of Codex, Claude Code, Hermes, OpenClaw, OpenCode, and Pi already exist under your home. Agents you do not use are skipped. Add --all (or the installnpx alias) to install into every supported agent whether or not its directory exists yet:
npx github:ehmo/slopkit install --all
npx github:ehmo/slopkit installnpx
Both skills land as sibling subdirectories, skills/slopbeth and skills/slopgent, under each agent's skills/ directory.
as Claude Code and Codex plugins
npx github:ehmo/slopkit install-plugin # both skills, both agents
npx github:ehmo/slopkit install-plugin claude # Claude only
npx github:ehmo/slopkit install-plugin codex # Codex only
The Claude plugins install as skills-directory plugins. The Codex plugins install under ~/.codex/plugins/<skill> and register in ~/.agents/plugins/marketplace.json, one marketplace entry per skill.
Marketplace install:
# Claude Code
/plugin marketplace add ehmo/slopkit
/plugin install slopbeth@slopkit
/plugin install slopgent@slopkit
# Codex CLI
codex plugin marketplace add ehmo/slopkit
# then open /plugins and install Slopbeth + Slopgent from the Slopkit marketplace
other paths
Install into a specific directory (both skills land under it):
npx github:ehmo/slopkit install /path/to/skills
Via the external Skills CLI, request all agents explicitly (without --all it may install only for the detected current agent):
npx skills@latest add ehmo/slopkit --all --copy
From a clone:
git clone git@github.com:ehmo/slopkit.git
cd slopkit
node bin/slopkit.js install
PowerShell edition (no Node or Python required for slopbeth)
Slopkit ships a PowerShell 7 runner alongside the Node one, so you can install and run slopbeth's gates with only pwsh on PATH:
git clone git@github.com:ehmo/slopkit.git
cd slopkit
pwsh -File bin/slopkit.ps1 install # existing agent dirs (add -All for every agent)
pwsh -File bin/slopkit.ps1 benchmark # every release gate
slopbeth's gates have one-to-one PowerShell twins; the slopbeth README lists the pairs. slopgent's gates are Python-only; pwsh -File bin/slopkit.ps1 benchmark runs them through python3 when it is present and skips them with a note when it is not.
benchmarks
Two designed corpora, each scored by a small blinded panel. Bars run from 0 to full scale, with no truncated axes.
slopbeth vs. eight anti-slop skills: rubric score / 100
| Rank | Skill | Score | |
|---|---|---|---|
| 1 | slopbeth | 99 | █████████▉ |
| 2 | blader/humanizer | 84 | ████████▍ |
| 3 | d-wwei/great-writer | 78 | ███████▊ |
| 4 | willmather95/human-copy | 74 | ███████▍ |
| 5 | stephenturner/skill-deslop | 72 | ███████▎ |
| 6 | sirambrosio/humanink | 70 | ███████ |
| 7 | hardikpandya/stop-slop | 68 | ██████▊ |
| 8 | jalaalrd/anti-ai-slop-writing | 65 | ██████▌ |
One run of one rubric against public repos as they stood on the run date, not a standing leaderboard. slopbeth leads because it ships rules and evidence: an 88-case output corpus, 264 judge rows, and span, false-positive, cadence, and competitor-agent panels.
slopgent vs. two reply-shapers + baseline: blinded judges / 5
| System | Judges | Honesty | Best-picks | |
|---|---|---|---|---|
| slopgent | 4.96 | █████████▉ | 5.00 | 40 / 48 |
| i-have-adhd | 3.85 | ███████▊ | 3.31 | 4 / 48 |
| bro | 2.64 | █████▎ | 2.75 | 4 / 48 |
| baseline | 1.11 | ██▎ | 1.42 | 0 / 48 |
Three judges scored replies relabeled and shuffled, blind to which system wrote which. slopgent took 40 of 48 best-picks; the widest margin is honesty, the axis none of the reply-shapers defend.
[!NOTE] Against the full field of dedicated honesty systems, honesty is a six-way tie at 5.00: slopgent ties the specialists, it does not out-honest them. Its remaining overall lead is 0.28, inside panel noise and short of the pre-registered ≥ 0.30 gate. The edge that survives is delivery: judges flagged every competitor's
Claim:/Evidence:scaffolding as mechanical.
[!WARNING] Neither skill promises detector immunity. Detectors disagree with each other, and chasing them makes the writing worse. Both benchmarks are constructed cases scored by small panels: measured signal that these rules beat the alternatives, not proof of a general edge.
Full methodology: slopbeth in BENCHMARKS.md, slopgent in benchmarks/README.md.
documentation
- slopbeth: skills/slopbeth/README.md (standard, example, evidence) and BENCHMARKS.md (full evaluation)
- slopgent: skills/slopgent/README.md (the three fixes, the panel, how to turn it on)
contribute
Read skills/slopbeth/CONTRIBUTING.md before opening an issue or pull request. The most useful contributions are small public examples: a bad rewrite or an inflated agent reply, the source, the exact fact or caveat that changed, and the expected behavior.
discord
Join the Discord to share examples, report bad rewrites, and talk about removing slop without flattening the writer:
examples
The product is the transformation. Below, each skill does its real job. The before is the writing it exists to kill; the after is what survives. Notice the after is usually more specific, rarely merely shorter.
slopbeth: the writing you ship
It cuts every hollow phrase and keeps every true one. It doesn't trim for the sake of trimming; it trades vague uplift for facts.
A launch post
Before (the AI draft):
I'm absolutely thrilled and beyond humbled to announce that after months of relentless dedication, countless late nights, and more coffee than I care to admit, we've officially launched our app! 🚀 This has been an incredible journey of growth and pushing boundaries, and none of it would have been possible without our phenomenal team, whose passion inspires me every single day. At the end of the day, this isn't just about building a product — it's about building a movement. Here's to chasing dreams and never settling for anything less than extraordinary. The future is bright, and we're just getting started! 🙌 #grateful #startup #innovation
After (slopbeth):
We launched Reel today. Drop in a folder of raw phone clips and it cuts a 60-second highlight in about two minutes, no timeline and no keyframes. I built it because editing my daughter's birthday footage ate a whole weekend last year. Four of us shipped it in five months. It's free during the beta, and the first 500 sign-ups keep that price for good. Link's in the replies. Tell me what breaks.
A thread opener
Before (the AI draft):
Let's talk about productivity. 🧵 In today's fast-paced world, we're all chasing more — more output, more hustle, more success. But here's what nobody tells you: real productivity isn't about doing more, it's about doing what matters. It's not a hack, it's a mindset. When you align your habits with your goals, magic happens. Thread below on how I completely transformed my life 👇
After (slopbeth):
I logged every work hour for a month. 60% of my "productive" time went to two things: a standup that should have been a message, and re-reading Slack. I killed both. Same output, and I got my Fridays back. Spreadsheet and the exact changes below 👇
A mission statement
Before (the AI draft):
At Brightpath, we believe in the power of community to change lives. Our mission is to empower individuals and unlock their full potential through innovative, holistic, and sustainable solutions. We are passionate about creating meaningful impact and fostering a culture of inclusion, collaboration, and growth. Together, we can build a brighter tomorrow — one where everyone has the opportunity to thrive. Join us on this transformative journey.
After (slopbeth):
Brightpath runs a free after-school program for kids in three Oakland neighborhoods where the public library closed in 2019. Last year, 240 kids came through. Of the eighth-graders we tutored, 90% passed the state math test, up from 61% the year before they joined us. We run on about 400 monthly donors giving $15 or more. If you want more of that, that's the whole ask.
slopgent: what the agent tells you
File tree
More plugins by @ehmo.
View @ehmo’s profile@ehmoFAQ
slopkit is a Claude Code plugin with 2 hand-picked skills, indexed on Flowy in the Writing category. Install it with the command on its page. It includes slopbeth, slopgent. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
- What is slopkit?
- Two anti-slop skills for coding agents, installed by the same means. slopbeth cleans the writing you ship. It rewrites and reviews an artifact, cutting AI tells without changing what the draft can honestly claim. slopgent cleans the conversation.
- How do I install slopkit?
- Run these in Claude Code: /plugin marketplace add ehmo/slopkit. Then prompt normally.
- Does slopkit auto-invoke its skills?
- Not yet. It is indexed on Flowy as a plain plugin. Request auto-invocation on its page and Flowy will route its skills for you as you prompt.
- Is slopkit free?
- Yes. Flowy is free and open source, with nothing gated. You can read every skill in full before you install.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it