mcpField
K
keep-it-human

keep-it-human

github.com/lelbornProductivity038 installsUpdated 1w ago
Repository

About keep-it-human

keep-it-human is a text-rewriting skill for authors who want more natural prose without changing its meaning or facts. It focuses on removing filler and common AI writing tics, and asks for a brief summary of the edits.

Key features

  1. 01Preserve facts, caveats, and intent
  2. 02Remove inflated wording and filler
  3. 03Vary sentence structure and rhythm
  4. 04Check formatting and common writing tics
  5. 05Include a short change summary

Use cases

  1. 01Polish prose that feels formulaic
  2. 02Retain caveats while rewriting text
  3. 03Adapt tone using an author's writing samples
  4. 04Find inflated or vague phrasing

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/lelborn/skills --skill keep-it-human

After installing, type/keep-it-humanin Claude Code to confirm it loaded.

Details

From the project README

Keep It Human

Rewrite text so it reads like a thoughtful person wrote it, not a language model — fixing the habits that make AI prose feel generic and instantly recognisable, while keeping the meaning, facts, and intent intact. The job is subtraction, not disguise.

The one rule that overrides the rest: fix the writing, never the substance. Don't drop a caveat, soften a claim, or invent a detail to smooth the prose. If a sentence is bad because the point is vague, that's a question for the author.

The tells

Think in clusters, not single instances — any one of these appears in good human writing too; it's the pile-up that signals a machine.

Inflated significance — hype the text hasn't earned: "stands as a testament to", "plays a crucial/pivotal role", "underscores the importance of", "in today's fast-paced/ever-evolving world", "in an increasingly digital landscape". And trailing "-ing" significance clauses: "..., highlighting the need for...", "..., showcasing its ability to..." — usually just delete them.

Overused vocabulary — one is fine, three in a paragraph is a giveaway: delve, tapestry, showcase, crucial, pivotal, realm, navigate, foster, robust, seamless, leverage, underscore, testament, boasts, elevate, myriad, landscape, ever-evolving, game-changer, unlock, harness, embark, resonate, meticulous, bustling. Use the plain word the author would.

Sentence-shape tics:

  • Forced rule of three: "fast, reliable, and scalable" — keep what matters, drop the item added for rhythm.
  • Negative parallelism: "It's not just X — it's Y", "This isn't about X. It's about Y." — say the positive thing directly.
  • False ranges: "from startups to enterprises" as filler.
  • Copula avoidance: "serves as / acts as / functions as" where "is" is meant.
  • Hollow "this": "This allows for", "This ensures that" pointing at nothing specific.

Formatting habits:

  • Em-dashes — the strongest surface tell; models overuse them badly. Cut hard (most become a comma, full stop, or parentheses). An occasional one is fine, but when in doubt remove it; same for en-dashes used as em-dashes.
  • Scattered bold on a phrase every other sentence — reserve it for the rare word that needs it.
  • Bullet lists where every item is **Bolded lead-in:** + clause — turn back into prose unless it's genuinely a list.
  • Heading tics: "X: The Complete Guide", Title Case On Everything, a decorative emoji per header — strip to plain sentence case.
  • Mismatched curly/straight quotes — make them consistent with the document.

Filler and hedging:

  • Generic conclusions: "In conclusion,", "Overall,", "Ultimately," restating what was said — cut, or end on something with content.
  • Hedge stacks: "It's important to note", "it's worth mentioning", "generally", "arguably" — one is honest, three is throat-clearing.
  • Chatbot residue: "Certainly!", "Great question", "I hope this helps!", "Let's dive in", "As an AI…", cutoff disclaimers, "Let's be honest,".
  • Both-sides padding: "While there are challenges, the future looks bright" — say the actual assessment.

Vague sourcing and invented specifics:

  • "Experts say", "studies show", "many believe" with nothing behind them — name the source or drop the appeal.
  • Plausible but unverified numbers, dates, or quotes the model may have invented — flag them for the author; never launder a fabrication into cleaner prose.

How to rewrite

  1. Read for meaning first — you're preserving that; everything else is negotiable.
  2. Rewrite for natural rhythm — vary sentence length (a short one after two long ones earns its keep), use contractions where the register allows, prefer concrete nouns and strong verbs.
  3. Audit against the tells — looking for surviving clusters, not lone instances.
  4. Final pass for em-dashes and stray tics — the cleanup that most changes how it reads.
  5. Summarise your changes in a line or two so the author can push back.

Match the author's voice

Given samples of their writing, study sentence length, vocabulary level, contractions, formality, and how they open and close, then rewrite toward that voice — the author on a good day, not a blander person. Without samples, aim for clear, plain prose at the formality the context calls for.

Don't overcorrect

  • Formal, polished, or technical writing isn't automatically AI — don't dumb down precise vocabulary or flatten sentences that earn their length.
  • Don't manufacture "humanity" with injected typos, forced slang, or random casual asides — that reads as fake.
  • A single em-dash, list, or formal word proves nothing; change things because they're padding or tics, not to hit a quota.
  • Keep every fact, caveat, and citation. Cleaner prose that says less is a worse result.

Before / after

Before: In today's ever-evolving digital landscape, leveraging robust automation tools has become a crucial component of any successful strategy — empowering teams to seamlessly streamline workflows, boost productivity, and unlock new opportunities for growth.

After: Automation tools have become part of how most teams work. They cut out repetitive steps, which frees people up for the work that actually needs a person.

Dropped: the hype opener, the overused vocabulary, the em-dash, and the forced triple — leaving the one concrete point.

Comments

Markdown supported