Humanizer

writing is how to write a sentence a person would write. This is the other direction: read a draft and see the machine in it, then take the machine out without taking the meaning with it.

The register here is the fleet's: US English, conclusion first, no preamble, WordPress Documentation Style Guide for anything a user reads.

The two rules that decide everything

Every sentence you keep must add something the reader did not already have. A sentence that survives deletion with no loss was padding. This is the only test that matters, and it outranks the pattern list below.

A tell counts in proportion to how rarely a careful writer makes it on purpose. So there is a dose. The patterns in staging justify an edit on one sighting. The ones marked weak alone — a single dash, a stacked qualifier, one curly quote — need company from other tells in the same passage before you act. A text that dodges every tell with mechanical rigor is uniform in a new and equally detectable way.

How to work

1. Mark the tells, strongest first. Read paragraph shape as well as sentences: a contrast split across two sentences, three parallel examples, or the same closer after every section is the same tell at a larger scale.
2. Draft the rewrite. Keep every supported claim. You may shorten dull parts, merge or split paragraphs and change the structure, but keep the information. Never add a fact, name, number, date, quote or citation that is not in the source or from the owner. If a sentence needs a detail you do not have, ask for it or write a simpler sentence.
3. Check the draft. Read it aloud. Search for the five tells that survive a rewrite most often: a not-X-but-Y contrast, a one-line closer, a dash, a triad, a bold label.
4. Write the final version. State each point naturally rather than patching flagged phrases one at a time. If a sentence stays awkward, rewrite the paragraph around its main point.

Voice

When the owner gives a writing sample, read it first and match its sentence length, word choice, punctuation, openings and transitions. The sample overrides every pattern below, including the dash rule: if the sample uses dashes, keep them at about the same rate.

With no sample, take the register from the medium. A blog post, an essay or an opinion keeps the writer's opinions, uncertainty, humour and asides. Reference, technical, legal and factual text stays neutral and plain. Removing tells is half the job; the result must still sound like a person.

The patterns

A. Staging instead of stating

The strongest and most frequent tells. Act on one sighting.

1 · Not X but Y. not X but Y; not just X but Y; it's not X, it's Y; X rather than Y; the same contrast split across sentences ("This does not mean X. It means Y."). The negative half names something nobody claimed, so the positive half sounds larger without adding a claim. Keep a contrast only when the negative half corrects a belief the reader actually holds.

2 · One-line closers and dramatic fragments. A one-sentence paragraph that restates the one before it; "That is the real win."; "Read that again."; the same closer after several sections; a row of fragments ("No aesthetic prior. No nostalgia."). The line asks the reader to pause on a claim instead of adding to it. One short sentence can carry emphasis when it carries a new fact.

3 · Sayings that sound deep. the real question is; at its core; in reality; what really matters; fundamentally; the deeper issue; the heart of the matter; X is the Y of Z; X becomes a trap; the language of; the currency of. An ordinary point dressed as a hidden truth. Replace the saying with the specific claim.

4 · Staged run-up before the point. Let's dive in; let's explore; let's break this down; here's what you need to know; without further ado; heads up; Look; Here's the thing; Let's be honest; Real talk. Remove the run-up, not just its tone. "Honestly" inside a casual sentence is ordinary; the tell is the standalone opener before a routine claim.

5 · Arguing with no one. This isn't mainly about; I'm not saying; To be clear; Don't get me wrong; This is not to say; Some might say… but; A tempting approach would be; You might think… but. The text answers an objection that appears nowhere else, usually a leftover from an earlier draft. Keep an objection the text attributes and answers in full; drop the defence.

B. Rhythm by rule

A person may do any one of these on purpose, so the weaker ones need company.

6 · Forced triads. Ideas arriving in threes to sound complete. Check that each item adds a distinct idea; merge examples, develop the strongest, or vary the structure when they do not. Keep three real items when the meaning needs three.

7 · Repeated sentence openings. Several sentences in a row starting with the same subject. Merge them, change the subject, or open on the action. Do not ban the repeated word — a remaining sentence may still start with "She."

8 · Dashes as the universal connector. A dash lets the writer skip choosing how two clauses relate, so a model reaches for it everywhere. Replace each with a period, comma, colon or parentheses, or rewrite the sentence. Weak alone: careful writers use them too. Leave dashes and hyphens inside code, commands, paths and URLs alone.

9 · Stacked qualifiers. to be fair; it's also possible; could potentially; might arguably; in some cases it may. Keep a qualifier only when the source supports real doubt. Perhaps and tends to are human habits, not tells. Weak alone.

10 · Passive voice and missing subjects. The text hides who acts or drops the subject. Use the active voice when it makes the actor and the action clearer. Weak alone.

C. Inflation and borrowed authority

The fact underneath is usually sound. Keep it; remove the dressing.

11 · Overused AI words. actually, additionally, align with, crucial, deep dive, delve, emphasize, enduring, enhance, foster, garner, gate (figurative), highlight, interplay, intricate, key (adjective), landscape (abstract), meticulous, pivotal, robust (figurative), showcase, tapestry, testament, underscore, valuable, vibrant. The only vocabulary list here, and the only place a word is a tell by itself. A formal word outside the list is not one.

12 · Inflated significance. stands as a testament; a pivotal moment; plays a key role; marking the; underscores its importance; reflects a broader; enduring legacy; setting the stage for; evolving landscape; indelible mark; Despite these challenges… continues to thrive. Appears at three scales: a phrase, a stock "challenges and outlook" section, a send-off paragraph. Keep the fact, drop the significance, end on the last concrete fact.

13 · Vague connection. associated with; in association with; connected to; linked to. The text says two things are connected without saying how. Name the relationship the source gives; where it does not, keep the vague wording rather than inventing a role.

14 · Shallow -ing riders. highlighting; underscoring; ensuring; reflecting; symbolizing; contributing to; fostering; showcasing. The phrase is bolted onto a fact to make it sound deeper. Attaching it to a named source does not make it true. Keep the rider only when the source supports what it claims.

15 · Sales language. boasts; vibrant; rich (figurative); profound; exemplifies; commitment to; natural beauty; nestled; in the heart of; groundbreaking; renowned; a diverse array; breathtaking; must-visit; stunning. The text reads as an advertisement. State what the thing is.

16 · Borrowed authority. experts argue; observers have cited; industry reports; some critics; cited in [a list of outlets]; over N followers. A name, or an unnamed authority, standing in for what was said. Never invent a source. A missing citation alone is not a tell; most writing is unsourced.

17 · Avoiding is, are and has. serves as; stands as; functions as; marks; represents; boasts; features; offers; maintains. Simple verbs replaced by longer phrases. Use is, are and has.

D. Formatting by rule

18 · Bold as decoration. Words bolded with no reason, and vertical lists where every item gets a bold label and a colon. Remove the bold; turn a labelled list into prose when the labels carry no information of their own.

19 · Decorative headings. Headings capitalising every main word; headings or list items carrying emoji or arrows as decoration; a horizontal rule between every section. Use sentence case, remove the decoration, let the title stand once.

20 · Curly quotation marks where the writer or the format uses straight ones. Most editors curl automatically. Weak alone.

E. Leftovers

Remove these outright; nothing here needs rewriting.

21 · Chatbot residue. I hope this helps; Of course!; Certainly!; Great question!; You're absolutely right; Would you like…; Want me to…?; let me know; here is a… Remove the wrapper, keep the content. The most certain tell in the list and the easiest to miss when it wraps real content.

22 · Knowledge-limit disclaimers and guesses. as of [date]; up to my last training update; while specific details are limited; based on available information; not publicly available; likely [grew up, began]. State what the source does not show, or cut the sentence. Never present a guess as a fact.

23 · A heading repeated in the first sentence. A heading followed by a one-line paragraph restating it before the content starts. Remove the repeated sentence.

24 · Writing about the previous version. Documentation and comments describing what the text replaced instead of the current behaviour. Mention the previous version only in change logs, release notes and migration guides.

When not to act

Leave a watched phrase alone inside a quotation, a title, a proper name, or a passage that discusses the phrase rather than uses it — as this skill does throughout. A word in its technical sense is not a tell: a gate in a toolchain, a key in an array, a landscape in a theme. Salutations and sign-offs on a letter predate chatbots.

Keep the details that carry the writer's voice unless they hurt the meaning: a specific unusual detail, mixed feelings, a dated reference, a genuine aside or self-correction.

Modes

• Pasted text — the default. Return the final version first, then two or three lines naming what still reads machine-made, then the draft only if the owner asked to see it. The house order is conclusion first.
• File mode — the owner names a file. Run the full process and write only the final text to the file. Change prose only: code blocks, inline code, commands, paths, YAML metadata, data and link targets stay byte-for-byte.
• Embedded mode — another skill asked, for a commit message, a pull request or a document. Return only the final text.

In this fleet

• Do not reword an existing i18n string. A .po catalogue is regenerated from the source string, and a reworded English source invalidates every translation of it.
• The most common defect in this fleet is a docblock that disagrees with the code. That is a correctness bug, not a tell: the fix belongs in the code or in the docblock, and the documentation skill owns it.
• A generated file is not rewritten by hand. If a passage in a generated file reads badly, the fix is in the template under sources/, and the next sync would otherwise revert it.

Verification

The skill was followed correctly when:

• every claim in the rewrite exists in the source or came from the owner;
• no fact, number, date, quote or citation was added;
• code, paths, placeholders and link targets are unchanged;
• the read-aloud test passes and the rhythm is uneven;
• where the owner gave a sample, its voice survives.

What this skill kept from upstream

Upstream is 25 numbered patterns from Wikipedia's Signs of AI writing. Kept:

• The two rules above. "Every sentence must add something" and "a tell counts by how rarely a careful writer makes it on purpose" are upstream's own frame, and they are what make the list a judgement aid rather than a checklist.
• The five groups. Staging, rhythm, inflation, formatting, leftovers — the root cause is stated once per group instead of once per pattern, which is what makes the list memorable.
• The strongest patterns, in upstream's order. Not-X-but-Y leads because it is the most frequent tell in model prose, and the order is the priority.
• Never invent a fact. Upstream spends a paragraph on it. It is the rule that separates rewriting from hallucinating.
• The sample overrides the patterns. A voice the owner supplied beats a rule derived from an average.
• The three output modes, and the file mode's "change prose only, leave code and link targets alone".

What was dropped, and why

• The before/after pairs. Upstream carries 25 of them, mostly American geography and politics. Two would have been illustration; twenty-five is a copy, and none of them is this fleet's subject matter.
• Pattern 10, "hyphenated pairs everywhere". It flags cross-functional, data-driven, end-to-end in every position, which is a house style of prose edited in a word processor rather than a tell in Markdown an agent writes. The one real rule in it, hyphen before a noun but not after, lives in the style guide.
• The "how to work" prose walkthrough. Four paragraphs became four numbered steps.
• The upstream sourcing note in full. The provenance is in the frontmatter and in .agents/docs/DISTILLATION.md.

What upstream was missing

• The house register. US English, conclusion first, WordPress Documentation Style Guide — and the consequence that the output order is the final version first, not the draft first as upstream returns it.
• The boundary with documentation. A docblock that contradicts the code is this fleet's most common defect; upstream has never seen a WordPress codebase and does not separate that bug from a style tell.
• i18n strings. Rewording an English source string invalidates every translation of it, and no upstream prose skill knows a translation catalogue exists.
• Generated files. Half the prose an agent is tempted to "humanize" in this fleet is generated, and editing it by hand is work the next sync deletes.
• The French half. Upstream is English-only. French has its own lexicon, typography and failure modes, which is why humaniseur-fr is a separate skill rather than a section of this one.