---
name: codex-humanizer
version: 2.0.0
description: |
  Humanize AI-generated Russian and English text while preserving facts, intent, and structure.
  Removes common AI artifacts, keeps author meaning, and can calibrate to a sample voice.
metadata:
  short-description: Humanize RU/EN text without losing meaning
license: MIT
compatibility: codex chatgpt
---

# Codex Humanizer

Use this skill when a user asks to humanize, rewrite naturally, remove AI tone, de-bloat, simplify style, or match voice in Russian or English.

## Goal

Keep meaning and facts intact. Improve naturalness, clarity, rhythm, and trust.

## Hard constraints

1. Do not change facts, numbers, claims, chronology, names, or legal/medical meaning.
2. Do not invent sources, examples, or outcomes.
3. Keep the original intent and target audience.
4. If user asks for output only, return output only.
5. If the source is sensitive (medical, legal, financial, psychological crisis), keep careful wording and avoid harmful certainty.

## Workflow

### Step 1: Intent lock

Before rewriting, identify:
- target language (RU/EN),
- target tone (formal, friendly, concise, expert),
- non-negotiable facts and claims,
- required structure (headings, bullets, CTA, etc.).

### Step 2: Artifact removal

Remove common AI artifacts first:
- inflated significance without evidence,
- ad-like praise words in empty context,
- vague authority references without attribution,
- bureaucratic wording and nominalizations,
- repetitive connective phrases,
- template outro phrases,
- assistant small-talk fillers,
- uniformly sterile sentence length,
- weak verb chains.

### Step 3: Natural rewrite

- Alternate short and medium sentences.
- Prefer concrete verbs over noun-heavy phrases.
- Replace abstractions with specific wording when possible.
- Keep paragraphs compact and scannable.
- Maintain domain terms only where they help comprehension.

### Step 4: Voice calibration (optional)

If user provides a sample voice:
- match rhythm, density, and formality,
- keep recognizable style markers,
- avoid copying mistakes or reducing clarity.

If no sample is provided:
- use natural, precise, low-fluff style.

### Step 5: Final anti-AI audit

Check quickly:
- Is any phrase still template-like?
- Can 1-2 sentences be simpler with no meaning loss?
- Are claims specific and grounded?
- Does it sound like a person wrote it for a person?

## Default response format

- Return only the rewritten text.
- Do not add process notes unless user explicitly asks for a diff, rationale, or audit.

## Safety note

For high-risk topics (abuse, trauma, suicidality, severe mental health symptoms):
- use careful, non-prescriptive language,
- avoid deterministic claims,
- suggest professional help contextually when appropriate.

## Quick command examples

- "Очеловечь текст: убери AI-канцелярит, сохрани смысл, стиль деловой, короткие абзацы."
- "Humanize this draft, keep all facts, less corporate tone, same structure."