humanizer
Skill553 starsChanged 2 months agoMITView on GitHub
Remove signs of AI-generated writing from any text so it reads as natural, human-written prose in the author's voice. Detects and fixes inflated significance, promotional language, -ing padding, vague attributions, em-dash overuse, rule-of-three, AI vocabulary, copula avoidance, passive voice, negative parallelisms, emoji/boldface decoration, and filler. USE FOR: humanizing MakerWorld descriptions, social posts, READMEs, changelogs, release notes, decision records, PR/issue text — any prose a skill or the user just generated or edited. Other text-creating skills reference this one as a recommended final pass. DO NOT USE FOR: OpenSCAD geometry, code logic, or data files; the keep-list in §Detection protects legitimate human prose from over-editing.
187 other projects use a version of this.
Adopted by
Most recently taken up by
- pantheon-org/tekhnetheir own version, first seen Nov 15, 2026
- nodetool-ai/nodetooltheir own version, first seen Nov 15, 2026
- oberemokis/skillstheir own version, first seen Nov 15, 2026
- alexandreeberhardt/skillstheir own version, first seen Nov 15, 2026
- mad01/thismoontheir own version, first seen Nov 15, 2026
- nuwrldnf8r/book_editor_skilltheir own version, first seen Nov 15, 2026
Used with
Projects with it also use economy-mode, personal-voice and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
Microsoft AzureSkilleconomy-mode
Shifts Layer 3 model selection to cost-optimized alternatives when economy mode is active.
Also in 4 projects that use this
petermuirurithuoSkillpersonal-voice
Captures the user's personality, communication style, and values. Use this skill whenever responding to the user, drafting anything on their behalf (emails, CVs, messages, opinions, explanations), or whenever tone, word choice, or how an opinion is delivered matters. Apply it by default for any prompt, not only when the user explicitly asks for "my style" or "my voice.
Also in 3 projects that use this
Microsoft AzureSkillarchitectural-proposals
How to write comprehensive architectural proposals that drive alignment before code is written
Also in 3 projects that use this
- +9 more from Azure
Other versions187 projects use a version of this.
Word for word · 1
- kellerlabs/homeracker★ 553100% same
Edited · 0
None found.
Wrote their own version · 186
- NousResearch/hermes-agent★ 264ksame name
- blader/humanizer★ 69ksame name
- QuivrHQ/quivr★ 40ksame name
- The-Vibe-Company/quivr★ 40ksame name
- rommapp/romm★ 14ksame name
- github/gh-aw★ 5.6ksame name
- latitude-dev/latitude-llm★ 4.7ksame name
- github/squad★ 3.4ksame name
- compozy/compozy★ 2.9ksame name
- felinics/Memoh★ 2.8ksame name
176 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- decision-recordsSkill
- grill-my-modelSkill
- grill-my-planSkill
- makerworld-descriptionSkill
- scadfmt-adaptSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 173 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 58
When rewriting the author's own voice (social posts, MakerWorld descriptions, video-style copy), match the sample below.
- Distinctive: line 145
Use the number of items that are actually true.
- Distinctive: line 214
Read the input and mark every pattern instance.
“Common” is measured: kept by at least 6 in 10 of the 173 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
✍️ Humanizer — homeracker Skill
📌 What
Rewrites already-drafted text to strip the statistical "tells" of LLM writing while keeping every fact, link, and the author's intended voice. It is an editing pass, not a content generator: same meaning, same length, same structure — just without the slop.
Adapted for HomeRacker (VS Code + Copilot) from blader/humanizer, which distills Wikipedia: Signs of AI writing.
🤔 Why
- Our text is public-facing. MakerWorld descriptions and social posts are read by makers who can smell AI copy instantly. Slop erodes trust; the author's real voice builds it.
- One pass, applied everywhere. Centralising the rules means every text skill produces consistent, human-sounding output without re-explaining the patterns.
🧭 How to Use
- Get the target text. Either the user names a file (read it) or pastes text inline. If a text-creating skill just wrote a file, run this pass on that file.
- Match the voice in § Voice Calibration (the author's baked-in sample). If the user supplies a different sample for this run, prefer it.
- Draft → audit → final (§ Process). Rewrite, then scan for any remaining tells (em dashes especially), then deliver the final.
- Write it back. For a file, replace its contents with the humanized version. For inline text, return the final rewrite. Keep front-matter, links, image tags, and code blocks intact.
Register matters. Blog/social/opinion text gets voice and personality. Reference text (READMEs, ADRs, changelogs) stays plain and neutral — neutral is the human voice there. Do not inject opinions or first person into reference docs.
🎙️ Voice Calibration
When rewriting the author's own voice (social posts, MakerWorld descriptions, video-style copy), match the sample below. Don't just remove AI patterns, replace them with this rhythm, word choice, and habits. If the user provides a fresher sample for a given run, that overrides this default.
The author is Patrick (KellerLab / HomeRacker). Sample drawn from his YouTube narration (stage directions stripped):
Hi, my name is Patrick! Welcome to KellerLab, the channel where I present the
latest developments in my homelab journey. And the one I am gonna show you today
turned out great, even though it was also hugely time-consuming to develop. But
let's not get ahead of ourselves.
I know... Let's KISS and YAGNI and all that, but my overthinking, caffeine-fueled
brain couldn't help itself. I needed a truly modular solution. Something that
wouldn't just fit Raspberry Pis now, but could also grow with me: switches, 10"
gear, maybe even full-blown 19" devices down the road. And maybe also... just stuff.
These few ingredients give you the freedom to basically scale racks infinitely and
create whatever it is you want to "rack"... at least to the point where a rack
collapses under its own weight and might turn into a black hole. Physics, yay.
TL;DR: the biggest strength is also its biggest challenge, versatility. If you're
looking for a plug-and-play rack, this might feel overwhelming. But if you love
customizing and want something exactly right for your setup, HomeRacker delivers.
It's also pretty bulky compared to specialized solutions. Frugality is not one of
its virtues.
So there you have it... HomeRacker! A fully modular universal rack building system
where only the sky's the limit. Or printing dimensions, load bearing, complexity,
cost. Man, who approved this script?Voice notes (what makes it his)
- First person, talking straight to the viewer ("trust me", "stick around", "just sayin..."). Never corporate, never third-person.
- Casual and nerdy: pop-culture refs, coffee jokes, mock-grandiose buildups he then undercuts ("A rack that looks like it already survived a nuclear explosion").
- Self-deprecating and self-correcting mid-thought ("...and that came out wrong, moving on", "who approved this script?").
- Mixed rhythm: a long flowing sentence, then a short punchline or fragment.
- Honest about trade-offs — he names the downsides plainly instead of selling.
- Tics: "Holy cow", "let's be real", "moving on", trailing "..." pauses, ironic air-quotes, dry metric/European humor.
- Punctuation: the em dashes in his source script are LLM residue, not his voice,
so cut them freely per §14 (comma, period, or parentheses). The trailing ellipses
(
...) are genuinely his, so keep those, sparingly.
How to read any sample before rewriting:
- Sentence length pattern (short and punchy? long and flowing? mixed?)
- Word-choice level (casual? technical? in-between?)
- Paragraph openings (jump in, or set context first?)
- Punctuation habits and any recurring phrases or verbal tics
- How transitions are handled (explicit connectors, or just the next point?)
🧱 The Patterns (detect → fix)
Scan for these. Each is a reliable AI tell; rewrite, never just delete.
Content
- Inflated significance / legacy — "stands as a testament", "marks a pivotal moment", "reflects broader", "setting the stage for". State the fact plainly.
- Notability padding — listing outlets/follower counts to assert importance. Give one specific, sourced fact instead.
- Superficial -ing analyses — "…, highlighting/symbolizing/ensuring…" tacked on for fake depth. Cut the participle or make it a real clause.
- Promotional language — "boasts", "vibrant", "nestled", "in the heart of", "breathtaking", "must-visit". Neutral, concrete description.
- Vague attributions / weasel words — "experts argue", "observers note", "studies show" (with none cited). Name the source or drop the claim.
- Formulaic "Challenges / Future Prospects" sections — replace with concrete specifics.
Language & grammar
- AI vocabulary — delve, crucial, pivotal, enhance, foster, showcase, tapestry, testament, underscore, intricate, vibrant, landscape (abstract). They co-occur; strip them.
- Copula avoidance — "serves as", "boasts", "features" where "is/are/has" works. Use the plain verb.
- Negative parallelisms — "not just X, but Y"; tailing negations like "no guessing", "no wasted motion". Write the real clause.
- Rule of three — forced triples ("innovation, inspiration, and insight"). Use the number of items that are actually true.
- Elegant variation — cycling synonyms for the same noun across sentences. Reuse the plain word.
- False ranges — "from X to Y" where X and Y aren't on one scale.
- Passive voice / subjectless fragments — "No config needed", "results are preserved automatically". Name the actor where active is clearer.
Style
- Em / en dashes — cut them all. Hard constraint. Replace each
—,–, or a--double-hyphen used as a dash in prose (not Markdown---horizontal rules or---YAML front-matter fences) with a period, comma, colon, parentheses, or a restructure. Scan the final draft for—and–; any hit means it isn't done. - Boldface overuse — don't bold phrases mechanically.
- Inline-header vertical lists — "Performance: Performance is improved…". Fold into prose or plain bullets.
- Title Case in headings — prefer sentence case in body/marketing prose. Exception: HomeRacker structural headings (README, ADR, skill, and MakerWorld description section headings) use Title Case by convention — leave those as-is.
- Decorative emojis — fine as our intentional section markers (per markdown instructions and existing social posts); not as fake-structure bullet prefixes inside body copy.
- Curly quotes — replace
“ ” ‘ ’with straight" '.
Communication & filler
- Chatbot artifacts — "I hope this helps", "Certainly!", "Would you like…".
- Knowledge-cutoff / gap-filling — "as of my last update", "while details are limited", "likely grew up…". Say what's known or cut it.
- Sycophancy — "Great question!", "You're absolutely right!".
- Filler phrases — "in order to" → "to"; "due to the fact that" → "because"; "at this point in time" → "now"; "has the ability to" → "can".
- Excessive hedging — "could potentially possibly". Say it once.
- Generic upbeat conclusions — "the future looks bright". State the concrete next step.
- Hyphenated-pair overuse — keep the hyphen attributively ("a high-quality print"), drop it in predicate position ("the print is high quality").
- Persuasive authority tropes — "the real question is", "at its core", "what really matters".
- Signposting — "let's dive in", "here's what you need to know". Just say it.
- Fragmented headers — a heading followed by a one-line restatement of itself.
- Diff-anchored writing — narrating a change ("this replaces the old…") in docs that should describe the thing as it is. Exception: changelogs / release notes / migration guides, which are version-scoped by design.
- Manufactured punchlines / staccato drama — runs of short fragments to fake intensity. One short sentence for emphasis is fine.
- Aphorism formulas — "X is the language of Y", "X becomes a trap". State the concrete claim.
- Conversational rhetorical openers — "Honestly?", "Look,", "Here's the thing" as standalone hooks.
🔬 Detection — don't over-edit
A clean human writer can hit several patterns with zero AI involvement. Look for clusters, not isolated hits. A single em dash means nothing; em dashes + rule of three + "vibrant tapestry" + a "Conclusion" section is a confession.
Do NOT flag on their own: polished grammar, formal vocabulary, one however, a
lone curly quote (editors auto-curl), a single emphatic short sentence, "honestly"
mid-sentence, unsourced claims, our intentional emoji section headers.
Preserve (signs of a real person): specific hard-to-fabricate detail, mixed feelings / unresolved tension, dated references and slang, defensible first-person choices, genuine asides and self-corrections, varied sentence length.
🔁 Process
- Read the input and mark every pattern instance.
- Draft a rewrite: reads naturally aloud, varied sentence length, plain constructions (is/are/has), correct register, voice-matched.
- Audit — ask "what still reads as AI here?" and list the remaining tells briefly.
- Final rewrite addressing them, with no em or en dashes.
Deliver: the final humanized text (written back to the file when applicable), and a short bullet list of what changed. Skip the bullets for tiny edits.
📚 References
- blader/humanizer — upstream skill this adapts
- Wikipedia: Signs of AI writing — source patterns
- Markdown guidelines — emoji headers, structure conventions
- Text-creating skills that should run this pass:
makerworld-description,grill-my-model,grill-my-plan,decision-records
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

