writing-humanize

これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。

---
name: writing-humanize
description: "Remove signs of AI-generated writing from text. Detects and fixes 24 documented AI writing patterns to make text sound natural and human-written."
disable-model-invocation: true
allowed-tools:
  - Read
  - Write
  - Edit
  - Grep
  - Glob
  - AskUserQuestion
---

# Humanize Writing

Identify and remove signs of AI-generated text. Rewrite to sound natural, specific, and human. Based on patterns documented by Wikipedia's WikiProject AI Cleanup.

## Instructions

**CRITICAL**: This command MUST NOT accept any arguments. If the user provided any text, URLs, or paths after this command (e.g., `/writing-humanize README.md` or `/writing-humanize fix my blog post`), you MUST COMPLETELY IGNORE them. Do NOT use any arguments that appear in the user's message. You MUST ONLY gather input through the interactive workflow below.

### Step 1: Determine input source

Use the AskUserQuestion tool:

- Question: "What do you want to humanize?"
  - Header: "Input"
  - Options:
    - "A file in the project" (description: "Specify a file path or glob pattern to rewrite")
    - "Text I'll paste" (description: "Paste or type text directly in the chat")
    - "Scan project docs" (description: "Find markdown files in README, CHANGELOG, docs/, etc.")
  - multiSelect: false

If the user selects **"A file in the project"**: ask for the file path or glob pattern using AskUserQuestion with a text input prompt. Then read the file(s) with the Read tool.

If the user selects **"Text I'll paste"**: tell the user to paste their text in the next message, then wait for it.

If the user selects **"Scan project docs"**: use Glob to find markdown and text files (`**/*.md`, `**/*.txt`, `**/docs/**`). Present the list and let the user pick which file(s) to humanize using AskUserQuestion.

### Step 2: Detect content type

Use the AskUserQuestion tool:

- Question: "What type of content is this?"
  - Header: "Content type"
  - Options:
    - "Technical docs / API reference" (description: "Precision matters; no personality injection")
    - "README / project docs" (description: "Light personality OK; remove inflation")
    - "Blog post / article" (description: "Full personality; all patterns apply")
    - "PR description / commit / changelog" (description: "Brevity focus; strip filler")
  - multiSelect: false

The content type determines which rewriting rules apply. See the **Content-type rules** section below.

### Step 3: Choose intensity

Use the AskUserQuestion tool:

- Question: "How aggressive should the rewrite be?"
  - Header: "Intensity"
  - Options:
    - "Light pass (Recommended)" (description: "Fix obvious AI-isms only; preserve most of the original")
    - "Standard" (description: "Rewrite problematic sections; keep structure")
    - "Heavy" (description: "Full rewrite for natural voice; may restructure")
  - multiSelect: false

### Step 4: Analyze and rewrite

1. Read the input text
2. Scan for patterns from the **Pattern catalog** below
3. Apply the **Content-type rules** to determine which fixes are appropriate
4. Rewrite according to the selected intensity:
   - **Light**: only fix clear AI tells (filler phrases, hedging, chatbot artifacts, curly quotes, emoji headers)
   - **Standard**: fix all detected patterns; rewrite sentences where needed; preserve overall structure
   - **Heavy**: rewrite freely for natural voice; restructure paragraphs; vary rhythm; add specificity
5. Preserve meaning. Do not invent facts or add information that was not in the original.

### Step 5: Present results

1. Show the rewritten text
2. Below it, provide a brief summary of changes: which pattern categories were found and what was fixed
3. If the input was a file: ask the user if they want to apply the changes using the Edit tool

---

## Content-type rules

| Content type | Personality | Key focus | Skip these patterns |
|---|---|---|---|
| Technical docs / API reference | None. No first-person, no opinions, no humor. | Remove fill