Humanizer vs Stop Slop: which removes AI writing tells?

Two free, open-source community skills, Humanizer by blader and Stop Slop by Hardik Pandya, take different approaches to stripping AI writing tells from prose: Humanizer rewrites AI-sounding text for rhythm and sentence variety, while Stop Slop applies seven rules to delete specific model-generated phrases and structures. Both are listed on SkillGild and installed via its CLI, and the common workflow is to run Humanizer first on a full draft and Stop Slop last as a final check. The projects' maintainers note the skills change style only and do not defeat AI-detection tools or verify facts.

Short answer: they do different jobs, and most people who care about this run both. Humanizer rewrites AI-sounding prose so it reads like a person wrote it, without changing what it says. Stop Slop is a shorter, stricter rule set that deletes the specific tells a model leaves behind. Humanizer changes how a draft moves; Stop Slop removes what should not be there. Running Humanizer first and Stop Slop last is the common order. Both are free, open-source community skills, listed on SkillGild and installed from their own repositories. Neither is ours. Humanizer https://skillgild.dev/skills/humanizer by blader rewrites AI-sounding text so it reads like a person wrote it, without changing what it says. It works from the patterns catalogued in Wikipedia's guide to AI writing, then audits its own output against the same list. It is the most-searched community skill of the year, which is a reasonable proxy for how widely the problem is felt. Stop Slop https://skillgild.dev/skills/stop-slop by Hardik Pandya is a skill file that teaches the model to recognise and remove its own writing tells: throat-clearing openers, not-X-but-Y contrasts, forced triads, filler closers. It is seven rules rather than a rewriting method, which is why it is quick and why it rarely changes your meaning. The distinction matters because the two failure modes are different. A draft can be free of every obvious tell and still read like a machine wrote it, because the sentences are all the same length and every paragraph has the same shape. That is Humanizer's problem to solve. Equally, a well-paced draft can still open with "In today's fast-paced world" and close with "the possibilities are endless". That is Stop Slop's. | | Humanizer | Stop Slop | |---|---|---| | Author | blader | Hardik Pandya | | What it changes | Rhythm, sentence variety, cadence | Specific phrases and structures | | Method | Rewrites, then audits its own output | Seven rules applied to the draft | | Risk to your meaning | Higher: it rewrites sentences | Lower: it mostly deletes | | Length of pass | Slower, it reworks the text | Fast | | Best used | First, on a full draft | Last, as a final check | | License | MIT | MIT | Both install the same way once the SkillGild CLI https://skillgild.dev/learn/install-claude-code-skills is set up: skillgild install humanizer --agent claude-code skillgild install stop-slop --agent claude-code Use --agent codex , cursor or gemini-cli for other clients. The order people settle on is draft, Humanizer, Stop Slop, read it yourself. Running Stop Slop first is not wrong, but it wastes work: Humanizer's rewriting can reintroduce a filler closer that Stop Slop had already removed. Be clear about the limit, because the category attracts overclaiming. These skills change prose style. They do not make text undetectable by AI-detection tools, and no skill honestly can: detection tools disagree with each other, change without notice, and produce false positives on human writing. If your reason for using one of these is to pass a detector, that is not a promise either project makes and not one we would repeat. They also do not check facts. A humanised paragraph with a wrong number in it is a wrong paragraph that reads nicely. If accuracy is the problem, that is a different pass. Writing Guidelines https://skillgild.dev/skills/writing-guidelines by Vercel is the third option worth knowing, and it solves a different problem again: it reviews prose against more than eighty rules from Vercel's style guide. It is built for documentation and product copy, so it is the right choice for help centres, onboarding emails and UI strings, and the wrong one for an essay. A fourth project, Caveman, takes the opposite approach: instead of smoothing prose it compresses it to terse, almost telegraphic output. It is popular and often mentioned alongside these two, but its repository restricts use of its name, so it is not listed on SkillGild and we link to it only in passing. Read its own repository and license before installing it, as you would with anything you find on GitHub. For the wider set, see the best Claude Code skills https://skillgild.dev/learn/best-claude-code-skills roundup, or best Claude skills for marketing https://skillgild.dev/learn/best-claude-skills-for-marketing if you are editing copy at volume and want the surrounding SEO and creative skills too. Both skills are open source under the MIT license, maintained in public repositories by named authors, and widely enough used that people search for them by name. We read each repository's README and SKILL.md before listing it, and both are installed from their own repositories rather than copied onto SkillGild. We have not run a controlled comparison of output quality between them, and this page does not claim one is better: they are different passes, and the sequence above is what their own documentation and common use suggest. Both appear in the catalog of Claude Code skills https://skillgild.dev/skills .