humanizer-ru
Skill13 starsChanged 56 days agoMITView on GitHub
Removes signs of AI generation from Russian-language blog content and social media posts (Telegram, vc.ru, Habr and similar) while keeping a lively, readable style. Universal and voice-neutral — works for long articles, short posts, notes, news breakdowns and opinion pieces. Must be used when the user says "перепиши более живо" (rewrite it more lively), "сделай менее ИИ-шно" (make it less AI-ish), "звучит как ChatGPT" (sounds like ChatGPT), "слишком сухо/шаблонно" (too dry / formulaic), "выглядит как LinkedIn-пост" (looks like a LinkedIn post), "очисти от воды" (cut the fluff). Also apply proactively after generating any Russian blog text, post or article longer than a paragraph, before handing the result to the user. Do not use for academic or legal texts — humanizer-ru-legal exists for those. Do not use for greentext, reply copy and other genres with their own conventions.
17 other projects use a version of this.
Adopted by
Most recently taken up by
- Vladimir-Human/humanizer-rutheir own version, first seen Nov 1, 2026
- jeremybrasher/grokbot-skillstheir own version, first seen Oct 25, 2026
- ilyautov/humanizer-rutheir own version, first seen Oct 19, 2026
- thevseprod/humanizer-rutheir own version, first seen Oct 9, 2026
- ponomarev-igor/humanizer-rutheir own version, first seen Oct 3, 2026
- karapuzpupkovich/humanizer-rutheir own version, first seen Sep 29, 2026
Other versions17 projects use a version of this.
Word for word · 1
- unicodef1wn/humanizer-russian-language★ 6100% same
Edited · 0
None found.
Wrote their own version · 16
- ilyautov/humanizer-ru★ 538same name
- smixs/humanizer-ru★ 146same name
- Vladimir-Human/humanizer-ru★ 141same name
- thevseprod/humanizer-ru★ 45same name
- comol/Humanizer_RU★ 16same name
- ivklgn/ai-kit★ 12same name
- ivklgn/bag-of-skills★ 12same name
- jeremybrasher/grokbot-skills★ 11same name
- KondrashovDenis/claude-humanizer-ru-skill★ 8same name
- BislanB/claude-skills-creativity★ 1same name
6 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 repository2 other files it gives its agents.
- editor-ruSkill
- proofreader-ruSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
23 rules. About 0 are common, 22 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈22
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
All examples and marker lists below are in Russian on purpose — they are the actual patterns the skill looks for in Russian text.
- Distinctive: line 33
Read the whole draft and catch the impression: where it sounds like a LinkedIn post, where like a ChatGPT essay
- Distinctive: line 36
Run the final audit (counters + structural checks)
- Distinctive: line 37
Make sure you haven't overshot into the opposite extreme (Section E)
- Distinctive: line 61
Fix: delete entirely. Start with a concrete fact, observation or claim about the topic.
- Distinctive: line 98
Fix: remove the mentor wrapper.
- Distinctive: line 120
Fix: delete. If content remains after deletion — keep it.
- Distinctive: line 136
Fix: delete entirely, leave the bare thought.
- Distinctive: line 223
Fix: remove or replace with a short "Но", "Однако", "Ещё", "Дальше", "А вот", "При этом".
- Distinctive: line 274
Fix: remove the summarizing phrases.
- Distinctive: line 292
Fix: if the items form a connected argument, rewrite as prose.
- Distinctive: line 294
Not causally linked to each other
- Very specific: line 305
If AI writes about N tools/methods/principles — each gets roughly the same amount of text.
- Distinctive: line 374
Fix: test every such phrase with the question "what exactly is being said here?".
- Distinctive: line 451
If you don't know the exact number — better to drop the evaluative word entirely than leave it vague.
- Distinctive: line 457
Check: every paragraph should have at least one very short sentence (3–7 words) and one longer one.
- Distinctive: line 517
Correct: remove the construction entirely or rephrase with an active verb.
- Distinctive: line 538
Correct: after cleaning the words, run the structure audit (B1–B6).
- Distinctive: line 592
Is the first paragraph specifics or pathos?: If "В современном мире..." — rewrite from a fact.
- Distinctive: line 596
Symmetry of coverage: if N points are listed, does each get the same volume?
- Distinctive: line 601
Run the reverse counters (Section E — "Protection against over-humanizing").
- Distinctive: line 605
Read the post aloud. Signs of finished text:
- Distinctive: line 636
All three paragraphs the same length — B1
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Humanizer for Russian blog content
The task is to remove signs of AI generation from Russian blog text (Telegram, vc.ru, Habr, articles, long posts, notes) while keeping a lively, readable style. This is not a "casual" rewrite down to the level of private messages; the goal is text that a competent author with their own voice could have written.
All examples and marker lists below are in Russian on purpose — they are the actual patterns the skill looks for in Russian text.
Scope
Covers:
- Long posts for Telegram channels (narrative format)
- Articles on vc.ru, Habr, Medium-like platforms
- Analytical breakdowns and reviews
- Personal opinion pieces and essays
- Tutorials and how-tos (the prose part, not the code)
Does not cover (different rules apply):
- Greentext posts
- Short reply copy (1–2 sentences)
- Video scripts (spoken rhythm is different)
- Academic texts, theses, legal opinions — use
humanizer-ru-legalfor those
Basic process
- Read the whole draft and catch the impression: where it sounds like a LinkedIn post, where like a ChatGPT essay
- Go through the marker checklist (Sections A–C)
- Rewrite the problem spots — don't swap individual words, restructure the phrases
- Run the final audit (counters + structural checks)
- Make sure you haven't overshot into the opposite extreme (Section E)
Core principle: rhythm, specificity and focus matter more than a replacement dictionary. You can remove every AI word and still leave the text dead if all sentences are the same length with no living connection between them.
Section A. AI markers in blog content (critical)
The most frequent markers by which a reader spots ChatGPT in a post within five seconds.
A1. Grandiose openers
The most recognizable AI marker in blogs. The text opens with a global statement instead of entering the topic.
Markers:
- "В сегодняшнем мире, где..."
- "В современную эпоху..."
- "В эпоху, когда технологии развиваются стремительными темпами..."
- "В условиях стремительно меняющегося..."
- "На фоне растущего интереса к..."
- "С развитием X / появлением Y..."
- "В последние годы наблюдается тенденция..."
- "Невозможно переоценить роль X в..."
Fix: delete entirely. Start with a concrete fact, observation or claim about the topic.
Before:
В современную эпоху, когда искусственный интеллект развивается стремительными темпами, всё больше компаний задумываются о внедрении AI-агентов в свои бизнес-процессы.
After:
AI-агентов сейчас внедряют все подряд. Большинство — впустую.
A2. The LinkedIn formula "here are N things I learned"
AI loves structuring a post as a motivational list with promised value.
Markers:
- "Вот 5 вещей, которые я понял о X"
- "3 урока, которые научили меня Y"
- "Делюсь опытом / лайфхаками / инсайтами"
- "Сохраняйте, пригодится"
- "Спасибо, что дочитали"
- "Если было полезно — ставьте 🔥"
- "Дайте знать в комментариях"
Fix: drop the formula. If the post really has N points — state them without the "here are 5 things" wrapper. Closing calls to action only if they are organic, not a ritual.
A3. Mentor tone
AI often writes from the position of a "life coach", even when discussing a technical detail.
Markers:
- "Важно понимать, что..."
- "Главное здесь — осознать, что..."
- "Запомните: ..."
- "Не забывайте, что..."
- "Помните: ..."
- "Усвойте одну простую мысль:"
- "Ключ к успеху в..."
- "Секрет в том, что..."
Fix: remove the mentor wrapper. If the thought is interesting, it works without "важно понимать". If it sounds banal without the wrapper — it was banal.
Before:
Важно понимать, что AI-агенты — это не магия, а просто инструмент.
After:
AI-агенты — не магия, а инструмент.
A4. Pseudo-depth tropes
AI pretends to be breaking through to the essence.
Markers:
- "по сути"
- "в сущности"
- "на самом деле"
- "если разобраться"
- "при ближайшем рассмотрении"
- "по большому счёту"
- "по факту"
- "истина в том, что"
Fix: delete. If content remains after deletion — keep it. If not — rewrite the whole phrase.
A5. Throat-clearing openers
AI almost never gets straight to the point.
Markers:
- "стоит отметить, что"
- "следует упомянуть"
- "необходимо подчеркнуть"
- "хочется отметить"
- "интересно отметить, что"
- "обращает на себя внимание"
- "заслуживает внимания тот факт"
- "не лишним будет сказать"
Fix: delete entirely, leave the bare thought.
A6. Bureaucratic register in a blog context
Normal in a thesis, an instant AI marker in a blog.
- "осуществляется" / "производится" → a concrete verb
- "является" → "это" / a dash / rephrase
- "данный" meaning "this" → "этот" or delete
- "в рамках" → can be removed in 80% of cases
- "в части касающейся" → delete
- "в целях" → "чтобы"
- "в случае если" → "если"
- "в условиях" → delete or make specific
- "механизм" meaning "how it works" → a concrete verb
- "функционирование" → "работа"
- "реализация" → "применение", "выполнение" or a verb
- "обеспечение" as a noun → a verb
- "требования, предъявляемые к" → "требования к"
- "осуществление деятельности по" → delete entirely, keep the verb
A7. Emotional inflation
AI inflates the importance and emotional charge of everything.
Markers:
- "это меняет всё"
- "революционный подход"
- "переворачивает индустрию"
- "не имеет аналогов"
- "впечатляющий результат"
- "поразительно"
- "уникальная возможность"
- "беспрецедентный"
- "невероятный"
- "потрясающий"
- "колоссальный"
Fix: replace with specifics or delete. If it really "changes everything" — explain exactly how, without the epithet.
Before:
Это революционная технология, которая меняет всё.
After:
Технология сокращает время разработки агента с двух недель до двух часов.
A8. Paired constructions
AI adores pairing.
- "как..., так и..."
- "не только..., но и..."
- "с одной стороны..., с другой стороны..."
- "и теоретически, и практически"
Fix: split into two sentences or list with commas.
Before:
AI-агенты полезны как для автоматизации рутины, так и для сложных задач.
After:
AI-агенты полезны и для рутины, и для сложных задач.
or:
AI-агенты автоматизируют рутину. С чем посложнее — тоже справляются, если правильно настроены.
A9. Participial tails
The sentence ends with a long "-ая/-яя/-ущ/-ющ/-вши" clause that adds nothing.
Markers: "подчёркивая", "отражая", "способствуя", "обусловливая", "формируя", "создавая", "представляя", "являясь", "обеспечивая", "позволяя"
Fix: cut the tail; if needed, move its content into a separate sentence.
Before:
Многие компании внедряют AI-агентов, стремясь автоматизировать рутинные задачи и сократить издержки.
After:
Многие компании внедряют AI-агентов — хотят сократить рутину и издержки.
A10. Bureaucratic bridges between paragraphs
AI links paragraphs with template bridges.
Markers: "между тем", "вместе с тем", "помимо этого", "наряду с этим", "в свою очередь", "в этой связи", "в этом контексте", "что касается", "переходя к"
Fix: remove or replace with a short "Но", "Однако", "Ещё", "Дальше", "А вот", "При этом".
A11. Upbeat ending
The most recognizable AI marker in a blog — a motivational-speaker finale.
Markers:
- "открывает новые горизонты"
- "знаменует новый этап"
- "будущее уже здесь"
- "перспективы безграничны"
- "пришло время действовать"
- "одно можно сказать точно: ..."
- "время покажет, но..."
- "впереди много интересного"
- "следите за обновлениями"
Fix: replace with a concrete conclusion or delete. A post can end on its last substantive point — closing pathos is not required.
A12. Hook formulas
AI has memorized blogger opening formulas and applies them every time.
Markers:
- "Я никогда не думал, что X. Пока не случилось Y..."
- "Вы наверняка слышали про X. Но мало кто знает, что Y..."
- "Что если я скажу вам, что..."
- "Представьте: ..."
- "Однажды я понял одну вещь..."
- "За последние N лет я сделал X. И вот что я понял..."
Fix: drop the formula, start with the content. If a hook is still needed — make it concrete (a number, an event, a name), not abstract.
Section B. Structural AI patterns
The hardest thing to catch is not individual words but the structure of the text.
B1. Symmetrical paragraphs
AI writes 3–5 paragraphs of roughly equal length (4–5 sentences each). A human author writes unevenly.
Fix: deliberately break the rhythm. One paragraph of two sentences, another of seven. Occasionally a single-sentence accent paragraph.
B2. Mandatory summary at the end of every paragraph
AI closes paragraphs with conclusions: "Таким образом...", "Следовательно...", "В итоге...", "Подводя итог...". A human author summarizes where it's needed, not after every paragraph.
Markers: "таким образом", "следовательно", "в итоге", "подводя итог", "резюмируя", "иначе говоря", "одним словом", "если коротко"
Fix: remove the summarizing phrases; let the paragraph end on its last substantive point.
B3. Duplicated intro and conclusion
AI opens the post with "what this will be about" and closes with "what this was about". Both parts paraphrase the main content.
Fix: compare the first and last paragraphs. If they say the same thing — delete one or rewrite it radically. In a blog post, no closing summary at all is better than a repeat of the intro.
B4. List mania
AI formats everything as bullets, even where prose is needed. This isn't structure, it's a marker.
Markers:
- A mandatory list in every section
- Lists of 3–5 items, each item 1–2 sentences
- Lists inside lists
- Lists where the items are full sentences with subject and predicate (meaning it was prose forcibly broken into bullets)
Fix: if the items form a connected argument, rewrite as prose. Keep lists only where the elements are:
- Genuinely enumerable (steps, options, tools)
- Not causally linked to each other
- Each readable on its own, without the neighbors' context
B5. Lists of exactly three
Almost every AI enumeration has three items: "X, Y и Z". Naturalness lies in irregularity.
Fix: sometimes leave two, sometimes four, sometimes a long dash-separated list.
B6. Symmetrical coverage of items
If AI writes about N tools/methods/principles — each gets roughly the same amount of text. A human author covers one in detail and dismisses another in half a sentence.
Fix: deliberate asymmetry — it reflects the real importance of the items.
Section C. Universal patterns (shared with the legal skill)
C1. Vague attributions
AI cites faceless "experts" and "the majority".
Markers:
- "эксперты считают"
- "по мнению многих"
- "большинство специалистов сходятся во мнении"
- "в индустрии принято считать"
- "общепризнанным является"
- "статистика показывает" (no number, no source)
- "исследования говорят" (no link)
Fix: either name the specific author/study, or rephrase without the appeal to authority.
Before:
Эксперты считают, что вайб-кодинг изменит индустрию разработки.
After:
Карпатый назвал это "vibe coding" в феврале 2025 года. С тех пор термин ушёл в массы.
or (if you don't remember the source):
Вайб-кодинг — это когда пишешь промпт, а ИИ генерит код. И с этим внезапно можно делать прод.
C2. Explanatory repetition
AI says the same thing twice in different words, "for clarity".
Before:
Агент работает автономно. Это значит, что он может выполнять задачи без вмешательства человека.
After:
Агент работает автономно — без человека.
C3. Explaining the obvious
AI explains what is already clear from context.
Before:
Claude, который является большой языковой моделью от компании Anthropic, может писать код.
After:
Claude умеет писать код.
A reader of an AI blog knows Claude is an LLM from Anthropic. If they don't — a separate sentence, not an inline insert.
C4. Safe generalizations
Formally true, substantively empty phrases.
Markers:
- "существуют различные подходы"
- "есть множество способов"
- "вариантов масса"
- "решения зависят от конкретного случая"
- "всё индивидуально"
- "у каждого свой опыт"
- "однозначного ответа нет"
- "тема сложная и многогранная"
Fix: test every such phrase with the question "what exactly is being said here?". If nothing — delete or replace with specifics.
C5. Synonym cycling
AI fears repetition and reaches for synonyms where a human author would calmly repeat.
Before:
Агент работает быстро. Этот инструмент справляется за минуты. Данное решение экономит время.
After:
Агент работает быстро — справляется за минуты.
C6. Dash mania and decorative em dashes
AI uses the long dash as decoration. For a human author the dash is a working mark.
Acceptable: replacing the copula "это"/"является", a strong pause instead of a comma.
Unacceptable: three dashes in one paragraph for rhythm, especially when they sit in the same position in every sentence.
C7. Excessive hedging
AI plays it safe:
- "может потенциально"
- "вероятно, что, возможно"
- "не исключено, что может быть"
- "в некоторых случаях иногда"
Fix: one hedge per unit of meaning, no more. In a blog it can often be removed entirely — the author's position sounds better without a hedge.
C8. Passive voice
AI-Russian overuses passive constructions.
Markers:
- "были выявлены / установлены / определены"
- "рассматривается как"
- "понимается как"
- "может быть охарактеризовано"
Fix: find who performs the action and make the construction active.
Before:
Агентами могут быть выполнены различные задачи.
After:
Агенты могут выполнять разные задачи.
In a blog the passive is almost always worse than the active — replace boldly.
Section D. Generative section — how to write well from the start
This is prevention, not cleanup. Write from these principles and there's less to clean later.
D1. Open with specifics
The first sentence of the post is not a global statement but a concrete fact, observation or claim about the topic.
Bad:
В современном мире AI-агенты становятся всё более популярными.
Good:
Anthropic выкатил Claude 4.7. Главное изменение — Claude Code теперь умеет работать с Excel.
D2. Specifics everywhere
Wherever you can put a number, a name, a date, a version — put it.
- "недавно" → "в марте 2026"
- "многие компании" → "Anthropic, OpenAI, Google"
- "значительный прирост" → "+40%"
- "несколько" → "три", "пять"
- "часто" → "в 70% случаев" (if you know) or delete
If you don't know the exact number — better to drop the evaluative word entirely than leave it vague.
D3. Rhythmic variety
Short sentences mixed with long ones. Occasionally a very short one as an accent ("И вот тут начинается интересное.").
Check: every paragraph should have at least one very short sentence (3–7 words) and one longer one.
D4. A position of your own
In a blog the author is entitled to a position. Not necessarily "я считаю" — it can come through:
- "Тут есть проблема: ..."
- "Эта штука переоценена."
- "Аргумент звучит убедительно, но..."
- "На практике это не работает, потому что..."
- "Решение прагматичное, но кривое."
AI is afraid to take a position. A human author isn't.
D5. Appropriate conversational turns (in moderation)
In blog text you may use (but not in every paragraph):
- "Беда в том, что..."
- "Получается, что..."
- "Главное здесь — ..."
- "Замысел понятен."
- "Это работает так:"
- "На пальцах:"
- "По факту..."
The norm is 2–3 per post, not per paragraph. Overdoing it makes the text "bloggerish" in the worst sense.
D6. Don't explain the obvious
Trust the reader. If they're reading a post about AI agents, they know what an LLM is. If the post is about vibe coding, they know Karpathy. Don't insert an explanation at every mention.
D7. Acknowledge complexity
Where there's nuance — mention it. AI often writes as if everything were unambiguous. A human author honestly says "тут спорный момент", "это работает не всегда", "у меня нет ответа".
D8. Concrete examples instead of abstractions
Back every abstract claim with an example. Not "AI-агенты экономят время" but "агент закрыл 30 заявок в Jira за час, что у меня заняло бы три дня".
Section E. What NOT to do (typical humanization mistakes)
Basic rules
- Don't turn it into TikTok copy. The goal is a lively blog style, not a 7-word post with emojis.
- Don't strip all structure. Headings and lists (where appropriate) are fine.
- Don't make every sentence short. Hemingway style in a blog is also a marker — not of AI, but of imitation.
- Don't insert "я" into every sentence. "Я думаю", "я считаю", "я заметил" — annoying after the third time.
- Don't use slang where it doesn't fit. vc.ru is not a school chat.
- Don't throw out technical terms. "AI-агент" stays "AI-агент", not "помощник на ИИ".
Humanization anti-patterns (bad edits)
Anti-pattern 1: Swapping in a filler synonym.
- "является" → "представляет собой" — same AI
- "следует отметить" → "стоит сказать" / "важно подчеркнуть" — filler synonyms
- "данный" → "указанный" / "рассматриваемый" — same bureaucratese
- "осуществляется" → "производится" / "реализуется" — just relocating the marker
Correct: remove the construction entirely or rephrase with an active verb.
Anti-pattern 2: Replacing passive with passive.
- "устанавливается" → "является установленным" — a different passive
- "применяется" → "находит применение" — worse than the original
Correct: find the subject and make the construction active.
Anti-pattern 3: Imitating a human author with a single device.
- "Беда в том, что..." everywhere — three times per post = a new marker
- A short accent sentence at the end of every paragraph — a recognizable pattern
- Dash inserts everywhere — looks like a mannerism, not living rhythm
Correct: variety of techniques, not repetition of one "humanizing" tell.
Anti-pattern 4: Cosmetics without structural editing.
Removing "следует отметить" from every paragraph — but leaving all paragraphs the same length with a summary at the end. The vocabulary is clean, the structure is AI.
Correct: after cleaning the words, run the structure audit (B1–B6).
Anti-pattern 5: Imitating a Telegram blogger.
- "короче" everywhere
- Ending sentences with "ну вы поняли"
- "лол", "кек", "топчик" every two paragraphs
In blog content this is no longer humanization but imitation of a specific genre. Don't do it if the task is a neutral blog style.
Anti-pattern 6: Switching to emotion.
Replacing AI inflation with your own emotional inflation: "это просто огонь", "это прям бомба". That's the other extreme.
Correct: specifics instead of emotion. "Это огонь" → "это сократило время с 8 часов до 30 минут".
Protection against over-humanizing (reverse counters)
| Marker | Acceptable | If more |
| conversational turns ("беда в том", "короче", "получается") | 2–3 per post | roll some back |
| sentences shorter than 5 words | up to 30% of all sentences | add connected reasoning |
| "я" / "мне кажется" / "по моему мнению" | in moderation, not in every sentence | make impersonal or remove |
| emojis | on request / by context, not as decoration | remove |
| exclamation marks | 0–2 per post | remove extras |
| rhetorical questions | 1–2 per post | remove extras |
| slang ("топ", "огонь", "база") | 0 in a neutral blog | replace with neutral |
Sign of over-humanizing: the post sounds like a schoolkid's Telegram channel rather than a vc.ru piece.
Final audit (mandatory step)
Counters (go through the text and count)
| Marker | Acceptable | If more |
| "следует отметить" / "стоит отметить" / "хочется отметить" | 0 | remove all |
| "является" | 1–2 per medium-length post | replace with a dash or a verb |
| "данный" | 0 | "этот" or delete |
| "осуществляется" / "производится" | 0 | a concrete verb |
| openers "в современном мире / эпохе" | 0 | delete |
| "важно понимать" / "запомните" / "главное" (mentor tone) | 0–1 per post | remove |
| summarizing paragraph endings ("таким образом", "в итоге") | 0–1 per post | remove extras |
| lists of exactly 3 items | no more than half | break the rhythm: 2 or 4 |
| paragraphs of equal length (±1 sentence) | no more than 2 in a row | split or merge |
| safe generalizations ("различные подходы", "есть много способов") | 0 | replace with specifics |
| emotional epithets ("революционный", "впечатляющий", "потрясающий") | 0–1 per post | replace with specifics |
| passive constructions | no more than 20% of sentences | make active |
| lists/bullets | only if the elements are genuinely enumerable | rewrite as prose |
Structural checks
- Is the first paragraph specifics or pathos? If "В современном мире..." — rewrite from a fact.
- Duplicated intro and ending: compare — if they say the same thing, delete one.
- Sentence rhythm: is there at least one short sentence (3–7 words) in every paragraph?
- Specifics: are there numbers, names, dates, versions in the text? If everything is at the level of general words — add specifics.
- Symmetry of coverage: if N points are listed, does each get the same volume? If yes — make it deliberately asymmetrical.
- Ending: does the post close with a grand phrase about "the future" or "prospects"? If yes — remove or replace with a concrete conclusion.
Protection against the reverse problem
Run the reverse counters (Section E — "Protection against over-humanizing"). If there's too much conversational tone, "я" or emotion — roll back part of the edits.
Final check
Read the post aloud. Signs of finished text:
- No "announcer's voice"
- Doesn't sound like a LinkedIn motivator
- Doesn't sound like a schoolkid's TikTok copy
- Has concrete facts, not only generalizations
- Has the author's position where appropriate
- Uneven rhythm — short and long sentences mixed
Worked example
Before (AI draft):
В современную эпоху, когда искусственный интеллект развивается стремительными темпами, всё больше компаний задумываются о внедрении AI-агентов в свои бизнес-процессы. Важно понимать, что AI-агенты представляют собой не просто инструмент автоматизации, а революционную технологию, которая открывает новые горизонты для бизнеса.
Существует несколько подходов к внедрению AI-агентов. Эксперты выделяют три основных направления: автоматизация рутинных задач, оптимизация процессов принятия решений и улучшение клиентского опыта. Каждое из этих направлений имеет свои особенности и требует индивидуального подхода.
Таким образом, AI-агенты — это технология, которая меняет правила игры. Будущее уже здесь, и компании, которые не успеют адаптироваться, рискуют остаться позади.
What's wrong:
- "В современную эпоху..." — A1
- "Важно понимать" — A3
- "Представляют собой" — A6 / anti-pattern (a synonym for "является")
- "Революционную технологию", "открывает новые горизонты" — A7, A11
- "Существует несколько подходов" — C4 (safe generalization)
- "Эксперты выделяют" — C1 (vague attribution)
- "Три основных направления" — B5 (rule of three)
- "Каждое имеет свои особенности и требует индивидуального подхода" — C4
- "Таким образом" — B2
- "Будущее уже здесь, рискуют остаться позади" — A11 (upbeat ending)
- All three paragraphs the same length — B1
After:
AI-агентов сейчас внедряют все подряд. Большинство — впустую.
Реальных сценариев, где агент окупается, всего три. Первый — рутина с чёткими правилами: разбор писем, обновление CRM, ответы на типовые вопросы. Это работает. Второй — задачи, где нужен поиск по большому объёму данных и быстрый ответ: агент копает базу знаний, человек принимает решение. Тоже работает. Третий — клиентская поддержка первой линии. Тут уже сложнее, и провалов больше, чем успехов: агент не понимает контекст, бесит клиентов, его приходится дублировать живым оператором.
Всё остальное — маркетинг. Если кто-то продаёт "AI-агента для стратегического планирования" — это либо обёртка над промптом в ChatGPT, либо чат-бот с тремя сценариями. Платить за это бессмысленно.
What was done:
- Removed the grandiose opener — the post opens with a concrete claim and a position
- Removed the mentor tone ("важно понимать") and the epithets ("революционная", "новые горизонты")
- "Эксперты выделяют" replaced with the author's own position
- "Три основных направления" — formally still three, but covered asymmetrically (first and second briefly, third with nuance)
- Paragraphs of different length: 2 sentences / 5 / 3
- Concrete examples instead of generalizations (CRM, knowledge bases, first-line support)
- Substantive ending, no "будущее уже здесь"
- The author's position and judgment appeared ("работает", "бесит клиентов", "бессмысленно")
Cheat sheet (top edits)
Always clean:
- "В современном мире / эпоху / в условиях X" → delete, start with a fact
- "Важно понимать / запомните / главное" → delete
- "Является X" → "X" / "— X" / rephrase
- "Представляет собой" → same thing, delete or rephrase
- "Данный" → "этот" or delete
- "Осуществляется" / "функционирование" / "механизм" → a concrete verb
- "Эксперты считают" / "большинство" → a specific source or your own position
- "Революционный" / "впечатляющий" / "меняет всё" → specifics
- "Таким образом" / "в итоге" at the end of a paragraph → delete
- "Будущее уже здесь" / "открывает горизонты" → a concrete conclusion or delete
- "Существуют различные подходы" → name which ones
- Prose lists → rewrite as prose
Never do:
- "Является" → "представляет собой" (filler synonym, same AI)
- "Беда в том, что..." everywhere (repeating a device = a new marker)
- Replacing AI inflation with your own ("это огонь", "это бомба")
Versioning
- v1.0.0 (current) — first version. Universal skill for blog content (Telegram, vc.ru, Habr and similar). Neutral voice. Adapted from
humanizer-ru-legalv1.2.0 + blog-genre specifics (LinkedIn formulas, mentor tone, emotional inflation, list mania, hook formulas, upbeat ending).
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.

