agentleFS
Sign inStart free
zdresearchzdresearch/p3n/skills/humanize/SKILL.md

humanize

Skill5 starsChanged 14 days agoMITView on GitHub

Rewrite a text paragraph by paragraph so it stops registering as AI-generated, without changing what it says or adding facts. Use when asked to humanize text, or to make AI-written text pass as human-written.

  • Installs packages

74 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use humanize-redo, humanize-korean 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.
  • dawn840705Skill

    humanize-redo

    가장 최근 윤문 결과를 2차로 다시 다듬는다 — 특정 카테고리·문단·강도 조정도 가능. humanize-korean 의 윤문 콜을 기존 run_id 에 재실행해 잔존 finding 을 처리한다. 트리거 — "$humanize-redo". **명시적 슬래시 호출 전용** — 자연어 후속 요청은 `humanize-korean` 오케스트레이터가 받는다.

    Also in 5 projects that use this

  • mskim8717Skill

    humanize-korean

    AI(ChatGPT·Claude·Gemini 등)가 쓴 한글 텍스트를 "사람이 쓴 글처럼" 윤문해주는 오케스트레이터 스킬. 번역투·영어 인용 과다·기계적 병렬·관용구·피동태 남용·접속사 남발·리듬 균일성·이모지/불릿 과다 등 10대 카테고리 40+ AI 티 패턴을 탐지·분류해 내용은 한 글자도 건드리지 않고 문체·리듬·표현만 자연스러운 한국어로 재작성한다. 트리거 — "AI 티 없애줘", "AI 같은 글 자연스럽게", "GPT/ChatGPT 문체", "AI 번역투 고쳐", "사람이 쓴 것처럼 윤문", "AI 윤문", "ChatGPT 티 제거", "한글 AI 탐지·윤문", "AI 글 사람처럼", "번역투 제거", "영어 인용 많은 글 윤문", "AI 글 티 안 나게", "휴머나이저", "humanize Korean", "AI detector bypass 한글". 후속 작업 — "특정 카테고리만 다시", "윤문 강도 조정", "장르 바꿔서", "이 문단만", "2차 윤문" 도 모두 이 스킬. 단순 맞춤법·오탈자 교정은 직접 처리, 번역은 번역 스킬, 내용 추가·삭제를 동반한 재작성은 별도 집필 스킬.

    Also in 3 projects that use this

  • CoralGameSkill

    perf-profile

    Structured performance profiling workflow. Identifies bottlenecks, measures against budgets, and generates optimization recommendations with priority rankings.

    Also in 3 projects that use this

  • Rayan55050Skill

    content

    Создание контента: посты, статьи, тексты

    Also in 3 projects that use this

  • luischanciSkill

    proofread

    Run the proofreading protocol on manuscript files. Checks grammar, typos, overflow, consistency, and academic writing quality. Produces a report without editing files.

    Also in 3 projects that use this

  • insight68Skill

    pricing

    Help figure out pricing for a product or service using minimalist entrepreneur principles. Use when someone is setting prices, considering price changes, or struggling with what to charge.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 64 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Oracle Check: If the AI detection oracle is not available here (see oracle section below) rewrite each paragraph once then go to step 4.

    line 20
  2. Very specific:

    If the user says yes, run the ai-detection sister skill on the result through its online hosted API (as there is no oracle on this machine).

    line 36
  3. Distinctive:

    If that cannot be installed here, the hosted online service can be used instead.

    line 43

“Common” is measured: kept by at least 6 in 10 of the 64 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Humanize

Rewrites each paragraph by rules generated via an AI detection oracle, and keeps the rewrite that scores least AI-like. For text the user wrote with a model and wants to read like human, the editorial sister skill is usually the better tool: it edits for plain meaning and asks for the facts a reader misses. Humanize changes how the text reads and registers, not what it says.

Inputs

  • Text: a file, or text the user provided.

It comes as the skill's argument (/humanize draft.md) or from the conversation.

Rewriters

Each rewrite is one subagent called rewriter, on the cheapest model you have (e.g., Luna for ChatGPT, Haiku for Claude), ideally at low effort. Tell it to read rules.md in this folder as its instructions; its message is one paragraph, and it answers with one JSON object: {"text": "the rewritten paragraph"}. Don't read rules.md yourself (context pollution): everything you need from it is here in this file. The rewrites are independent, you can run them in parallel.

The flow

  1. Pick the paragraphs. Rewrite prose paragraphs of 8 words or more. Headings, lists, tables, quotes, code and shorter lines (a sign-off, a label) stay as they are.
  2. Oracle Check. If the AI detection oracle is not available here (see oracle section below) rewrite each paragraph once then go to step 4.
  3. Rewrite. rewrite each of the first 5 paragraphs three times, once with each message below; rewrite any paragraph after the 5th once, with message A.
    • A: <Paragraph Text>
    • B: Apply every rule strictly, rules 7 to 12 above all: no lead-in clauses, no contrast pairs, no trailing which/since clauses, no topic sentence. Paragraph: <Paragraph Text>
    • C: Rules 1, 3 and 4 above all: drop every general statement, concrete part first, short sentences. Paragraph: <Paragraph Text>
  4. Guard the facts. If a rewrite has a number, or a capitalised word in mid sentence, that its paragraph does not have, ask that rewriter once more, adding "(Your last rewrite added X, which the paragraph does not contain. Do not.)". If it adds one again, keep the original paragraph.
  5. Pick the best. With three rewrites per paragraph (i.e., with the oracle), write each to a file and score them all in one call: python3 scripts/oracle.py FILE1 FILE2 .... Keep the one with the lowest ai.
  6. Put the text back together, one paragraph for one paragraph, with the user's layout between them. If the oracle is available, score the original and the result: python3 scripts/oracle.py BEFORE AFTER.

Show the user the text, the score before and after (a result passes when its top label is not ai), and any paragraph kept as sent because its rewrites added facts. Filler is cut on purpose, so tell the user to check that every claim they need is still there.

Without the oracle, end your reply to the user with this note, after the text (in the reply only, not in the text or its file):

Humanized without a quality check. The AI detection oracle isn't available on this system, so this rewrite wasn't scored and may still read as AI-generated. For a checked result, the oracle needs to be installed (or the hosted online API can be used). Do you want me to check online whether it still reads as AI?

If the user says yes, run the ai-detection sister skill on the result through its online hosted API (as there is no oracle on this machine).

Oracle

The quality and sanity check of humanization is dependent on an oracle AI detection classifier.

The oracle needs PyTorch and Hugging Face Transformers (pip install -r requirements-model.txt installs them), a 1.7 GB model download on first use, and about 2.2 GB of RAM; no GPU required. On CPU it takes about 5s to load, then about 2s per paragraph check.

If that cannot be installed here, the hosted online service can be used instead.

If the oracle is installed and available, python3 scripts/oracle.py FILE gives a score. Each call loads the model fresh (about 5s), and files given together share one load. To score in parallel or many times, start it once with python3 scripts/oracle.py --keep-alive: later calls use the loaded model, at one model's memory however many run at once. Stop it with python3 scripts/oracle.py --stop when done; it also unloads itself after 10 minutes without work.

Hosted API

Only if you can't spawn subagents, or the user wants the scored version and the classifier doesn't run here: the hosted p3n API runs the same rules with the classifier. Read https://p3n.si/skills/humanize#for-ai-agents then, and follow it; it has the calls, the key and the free limit.

The user's p3n API key should live in ~/.config/p3n/key.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free