ai-writing-detection

name

ai-writing-detection

description

Detects AI-generated text and rewrites/de-AIs documents to sound natural and human. Use this skill whenever the user wants to: check if writing sounds AI-generated, scan text for AI tells, clean up AI writing, humanize a document, remove AI patterns, rewrite a doc to sound less like ChatGPT, or de-AI any text. Triggers on phrases like "does this sound AI?", "humanize this", "de-AI this", "rewrite this to sound human", "check for AI patterns", or whenever a file or text is provided alongside intent to detect or fix AI-like writing.

AI Writing Detection & Rewriting

Two modes, one skill:

• Detect — Scan text and report AI patterns found
• Rewrite — Rewrite the text to remove AI patterns, preserving meaning and voice
• Both (default when no flag) — Detect first, then rewrite

If the user says "de-AI this", "humanize this", "rewrite this", or provides a doc/file and asks to fix it → use Rewrite mode.
If the user says "check this", "does this sound AI?", "scan for AI tells" → use Detect mode.
If unclear → run Both.

---

Step 1: Read the AI Pattern Reference

Before running either mode, read the full pattern list from:

→ references/ai-patterns.md

This contains the canonical list of words, phrases, and structural patterns that signal AI writing. Load it into context before analyzing or rewriting anything.

---

Step 2: Detect Mode

Scan the text against the patterns in references/ai-patterns.md.

Output format

## AI Writing Detection Report

**Overall assessment:** [Clean / Minor AI signals / Moderate AI signals / Strong AI signals]

### Em Dashes
[Count found] — [list each instance]

### Flagged Words & Phrases
| Found | Category | Suggested replacement |
|-------|----------|-----------------------|
| ...   | ...      | ...                   |

### Structural Patterns
- [Any structural patterns found, e.g. "Whether you're a X, Y, or Z..."]

### Summary
[2–3 sentence plain English summary of what was found and how noticeable it is]

Scoring guide:

• 0 flags → Clean
• 1–3 flags → Minor AI signals
• 4–8 flags → Moderate AI signals
• 9+ flags → Strong AI signals

---

Step 3: Rewrite Mode

Rewrite the text to eliminate AI patterns while:

1. Preserving meaning — Every idea in the original stays in the rewrite
2. Preserving structure — Don't restructure paragraphs unless absolutely needed
3. Preserving voice — Match the formality and tone of the original
4. Not over-correcting — Don't replace every word; only fix flagged patterns
5. Not introducing new AI patterns — Check your own output before finishing

Rewriting rules

• Replace flagged words with simpler, more direct alternatives (see references/ai-patterns.md)
• Convert em dashes to commas, parentheses, or colons as appropriate
• Break up repetitive sentence structures if found (vary length naturally)
• Remove filler intensifiers (absolutely, truly, certainly, etc.) unless removing them breaks grammar
• Do NOT add new content, examples, or elaboration — rewrite only what's there
• Do NOT change proper nouns, technical terms, or quoted text

Output format

Return the full rewritten text, then a brief change log:

[Rewritten text here — complete, ready to use]

---
**Changes made:**
- Replaced "leverage" → "use" (×2)
- Converted 3 em dashes to commas
- Removed "robust", "comprehensive" → "thorough", "complete"
- Removed filler: "truly", "essentially"

---

Step 4: File Handling

If the user provides a file:

• .txt / .md — Read with bash cat, detect/rewrite, output as same format
• .docx — Read the skill at /mnt/skills/public/docx/SKILL.md before proceeding. Extract text, rewrite, then write back as .docx and present the file.
• .pdf — Read the skill at /mnt/skills/public/pdf-reading/SKILL.md. Extract text, rewrite, present as .txt or .md (PDF regeneration is complex — ask user their preferred output format)
• pasted text — Process inline, show output in conversation

---

Quick Reference: Top AI Signals

The most common and detectable patterns (full list in references/ai-patterns.md):

#1 tell: Em dashes (—) used more than once per page
Top verbs: delve, leverage, utilise, facilitate, foster, underscore
Top adjectives: robust, comprehensive, pivotal, seamless, nuanced, transformative
Top openers: "In today's...", "It's important to note...", "In the realm of..."
Top closers: "In conclusion...", "At the end of the day...", "By doing X, you can Y..."
Filler words: absolutely, essentially, truly, significantly, undoubtedly