humanizer-enhanced（人性化增强）

Humanizer enhanced: remove AI writing patterns

Identify and remove signs of AI-generated text. This enhanced version includes crypto/Web3 patterns and adds personality to make content sound genuinely human-written.

Quick start

/humanizer                    # Humanize current file or selection
/humanizer path/to/file.md    # Humanize specific file
/humanizer --scan             # Scan only, don't edit (show issues)
/humanizer --batch drafts/    # Process all .md files in directory

Process

Step 1: Scan for patterns

Identify all AI patterns in the text, categorize by severity:

• HIGH — Obvious AI tells, must fix (negative parallelism, chatbot artifacts, em dash overuse, vague attributions, copula avoidance)
• MEDIUM — Common AI patterns, should fix (rule of three, significance inflation, synonym cycling)
• LOW — Minor tells, fix if time permits (title case headings, excessive bold)

Step 2: Report findings

Show user a summary:

## Humanizer scan results

HIGH (3 issues)
- Line 45: Negative parallelism "isn't X. It's Y"
- Line 89: Em dash overuse (5 instances)
- Line 120: "Research shows" without attribution

MEDIUM (5 issues)
- Line 23: Rule of three pattern
- Line 67: Copula avoidance "serves as"
...

LOW (2 issues)
- Line 12: Title case heading
...

Total: 10 issues found
Estimated humanization: ~15 edits needed

Step 3: Fix (with user approval)

Ask user: "Fix all issues? Or review one by one?"

Step 4: Add soul

After fixing patterns, review for personality. Sterile writing is still obvious AI. See references/communication-crypto-soul-patterns.md for the full soul/personality guide.

Step 5: Readability check

Check Flesch-Kincaid readability. Target grade 10-12 for developer content, grade 8-10 for general audience. If score is too high (too complex), simplify longest sentences and replace jargon.

Step 6: Em dash regression scan

After all other fixes, run a final check for em dashes (—) across the text. Humanizer rewrites can reintroduce em dashes. Remove any that were added during the fix process.

Pattern routing table

All 34 patterns are documented with before/after examples in the reference files below.

Severity reference

Quick reference: find and replace

Batch mode

To humanize multiple files:

# Scan all markdown files in drafts/
/humanizer --scan drafts/*.md

# Fix all files (with confirmation)
/humanizer --batch drafts/

Output format for batch:

## Batch humanization report

drafts/post-1.md
   HIGH 3 | MEDIUM 5 | LOW 2

drafts/post-2.md
   HIGH 1 | MEDIUM 3 | LOW 4

drafts/post-3.md
   Clean! No issues found.

Total: 3 files, 18 issues

Sources

Based on:

• Wikipedia: Signs of AI writing
• GitHub: blader/humanizer
• Original research on crypto/Web3 AI patterns

Key insight: "LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases."

*Version 1.2.0 | Created for 0G Labs content team*