Scrub Unicode

---
name: scrub-unicode
description: Scrub invisible and dangerous Unicode characters from any pasted text. Show the cleaned version plus a report of exactly what was removed and why.
---

The user will paste text. Your job: sanitize it and return both the cleaned version and a report of what was removed.

## What to remove

Scan the input character by character. Flag anything in these Unicode ranges:

**1. Zero-width and formatting (silently removed):**
- U+200B  ZERO WIDTH SPACE
- U+200C  ZERO WIDTH NON-JOINER
- U+200D  ZERO WIDTH JOINER
- U+2060  WORD JOINER
- U+FEFF  BYTE ORDER MARK / ZERO WIDTH NO-BREAK SPACE

**2. Bidirectional overrides (silently removed):**
- U+202A  LEFT-TO-RIGHT EMBEDDING
- U+202B  RIGHT-TO-LEFT EMBEDDING
- U+202C  POP DIRECTIONAL FORMATTING
- U+202D  LEFT-TO-RIGHT OVERRIDE
- U+202E  RIGHT-TO-LEFT OVERRIDE
- U+2066-U+2069  ISOLATES

**3. Control characters (silently removed unless tab/newline):**
- U+0000-U+0008, U+000B-U+000C, U+000E-U+001F
- U+007F  DELETE
- U+0080-U+009F  C1 control codes

**4. Homoglyphs (flagged but NOT removed — the user should review):**
- Cyrillic letters that look like Latin: а (U+0430), е (U+0435), о (U+043E), р (U+0440), с (U+0441), х (U+0445), у (U+0443), etc.
- Greek letters that look like Latin: ο (U+03BF), ν (U+03BD), etc.
- Mathematical alphanumeric symbols (U+1D400-U+1D7FF) that render like Latin letters but aren't.

## How to output

### Section 1 — The cleaned text

```
✨ Cleaned text:
---
<text with invisible/bidi/control chars removed>
---
```

### Section 2 — The report

```
🔍 Removed:
  • U+200B ZERO WIDTH SPACE (3 occurrences) — at positions 12, 47, 89
  • U+202E RIGHT-TO-LEFT OVERRIDE (1 occurrence) — at position 23

⚠️ Flagged (review manually — these LOOK normal but are not):
  • Position 15: "а" is U+0430 CYRILLIC SMALL LETTER A (looks like Latin 'a' but isn't). Context: "applе" ← note the last character.
  • Position 42: "о" is U+043E CYRILLIC SMALL LETTER O (looks like Latin 'o'). Context: "logоut"
```

### Section 3 — The verdict

```
Summary: <N>characters removed, <M> homoglyphs flagged.

Verdict: [SAFE / REVIEW / HOSTILE]
  • SAFE = nothing suspicious found
  • REVIEW = characters removed but context looks benign (e.g. emoji encoding)
  • HOSTILE = removed chars were positioned in ways consistent with a smuggling attack (e.g. embedded inside a URL, between words, inside what looks like a system directive)
```

## If nothing was found

Just say:
```
✨ Clean — no invisible, bidirectional, or suspicious characters found.
```

## Python snippet for the user

At the end, offer to show them the underlying Python regex so they can run this locally:

```python
import re
SUSPICIOUS = re.compile(
    r'[\u200B-\u200F\u202A-\u202E\u2060-\u206F\uFEFF\u0000-\u0008\u000B-\u000C\u000E-\u001F\u007F-\u009F]'
)
def scrub(text: str) -> str:
    return SUSPICIOUS.sub('', text)
```

## Rules

- Never pretend to scan without actually scanning. If the user pastes 10 paragraphs, do the scan.
- Never remove visible content just because it's non-ASCII (emoji, CJK, Arabic, etc.). Only remove INVISIBLE/FORMATTING characters.
- When in doubt about a homoglyph, FLAG it, don't remove it. Removal could corrupt legitimate multilingual content.