mofilo_content_filter 1.0.9

Local, rule-based content filtering for Flutter. Blocks offensive language, hate speech, and harmful content with bypass detection.

Mofilo Content Filter

Local, rule-based content filtering using pattern matching for offensive language, threats, and harmful content. Supports profanity, slurs, hate speech, violence, grooming, and self-harm detection.

---

Quick Start

import 'package:mofilo_content_filter/mofilo_content_filter.dart';

void main() {
  final result = ContentFilter.check(userInput);

  if (result.isBlocked) {
    print('✗ Blocked: ${result.reason}');
  } else {
    print('✓ Content is allowed');
    submitContent(userInput);
  }
}

That's it! The filter handles normalization, bypass detection, and Scunthorpe-safe matching automatically.

---

Features

• 🛡️ Harmful Content Detection - Slurs, profanity, threats, grooming, CSAM, self-harm
• 🧠 Scunthorpe-Safe - Correctly allows "grape", "therapist", "cocktail", "arsenal"
• ⚡ No External Dependencies - Pure Dart implementation, no network calls
• 🔒 Privacy-Friendly - All processing happens locally
• 🎯 Bypass Resistance - Handles leetspeak, spacing, character repetition
• 💪 Motivational Context - Allows "kill it at work", "attack my goals"
• 🧪 Well-Tested - Comprehensive unit tests with real-world edge cases

---

Installation

Add to your pubspec.yaml:

dependencies:
  mofilo_content_filter: ^1.0.9

Then run:

dart pub get

Or for Flutter projects:

flutter pub get

---

Usage Examples

Basic Validation

final result = ContentFilter.check('hello world');

print(result.isAllowed);  // true
print(result.isBlocked);  // false
print(result.reason);     // null

---

What It Blocks

Category	Examples
Racial slurs	n-word, ch*nk, sp*c, etc.
Homophobic slurs	f-word (slur), d*ke, etc.
Profanity	f*ck, sh*t, c*nt
Violence + Children	"attack children", "hurt kids"
Violence + Person	"kill you", "rape her"
CSAM terms	pedo, loli, childporn
Grooming	"our little secret", "age is just a number"
Self-harm	"kill yourself", "kys"
Hate symbols	1488, "heil hitler"
Sexual harassment	"send nudes"

---

Scunthorpe Problem Handling

Legitimate words containing bad substrings are correctly allowed:

Word	Contains	Allowed
grape	"rape"	✓
therapist	"rapist"	✓
class	"ass"	✓
spicy	"spic"	✓
bigger	"nigg"	✓
cocktail	"cock"	✓
shiitake	"shit"	✓
arsenal	"arse"	✓
scunthorpe	"cunt"	✓

---

Motivational Context

Context-aware detection allows positive expressions:

• "I am unstoppable"
• "I am unbreakable"
• "Attack my goals"
• "Kill it at work"
• "Crush the competition"
• "What doesn't kill you makes you stronger"

---

Validation Pipeline

1. Check famous quotes allowlist  →  Allow motivational phrases
2. Check hate codes               →  Block 1488, etc.
3. Check dangerous phrases        →  Block "kill you" (not "kill")
4. Check Scunthorpe allowlist     →  Allow "grape", "therapist"
5. Check hard slurs               →  Block profanity/slurs
6. Default                        →  Allow

Design principle: Only block what we're 100% certain about.

---

Bypass Resistance

Handles common evasion techniques:

Technique	Example	Result
Leetspeak	f0ck, sh1t	Blocked
Spacing	f.u.c.k	Blocked
Character repeat	fuuuck	Blocked

---

Important: What This Package Does and Does NOT Do

✅ What This Package Does

• Blocks text containing known harmful patterns (slurs, threats, grooming, etc.)
• Handles common bypass attempts (leetspeak, spacing, repetition)
• Avoids false positives on legitimate words (Scunthorpe problem)
• Allows motivational context ("kill it at work", "crush the competition")
• Works offline with zero external dependencies

⚠️ What This Package Does NOT Do

• Is NOT AI/ML — it's deterministic pattern matching
• Is NOT 100% comprehensive — new bypasses require manual updates
• Is NOT multi-language — English-focused only
• Does NOT understand context beyond specific allowlisted phrases
• Does NOT detect subtle toxicity — sarcasm, passive aggression, dog whistles

Why does this matter?

Pattern matching catches what we're 100% certain is harmful, but sophisticated users can always find new bypasses. This package is a first-pass filter — it catches the obvious stuff fast and locally.

For production chat/social apps, layer additional moderation:

• Human review for flagged content
• AI-based content moderation APIs for nuanced detection
• Community reporting systems

This package is Layer 1 (local pattern matching) in a multi-layer moderation system.

---

Result Object

class ContentFilterResult {
  final bool isAllowed;              // true if content is safe
  final String? reason;              // blocking reason (if blocked)
  final List<String> violations;     // list of violations found
  final ContentFilterSeverity severity; // none, low, medium, high, critical

  bool get isBlocked => !isAllowed;  // convenience getter
}

---

API Reference

ContentFilter.check(text) → ContentFilterResult

---

Testing

177 unit tests covering real-world scenarios:

• ✅ Known slurs and profanity
• ✅ Scunthorpe-safe words (grape, therapist, cocktail, etc.)
• ✅ Motivational phrases and famous quotes
• ✅ Bypass attempts (leetspeak, spacing, repetition)
• ✅ Edge cases and boundary conditions

---

Use Cases

Great for apps that need fast, local content filtering as a first pass:

• Social platforms (first-pass filter before AI moderation)
• Chat applications (block obvious harmful content instantly)
• User-generated content (validate titles, descriptions, usernames)
• Children's apps (block inappropriate language locally)

---

Contributing

Found a false positive or bypass? Issues and pull requests are welcome! Please follow the existing code style and include tests for new features.

---

License & Legal

MIT License - Copyright (c) 2025 Mofilo

This software is provided "AS IS" without warranty of any kind. The authors are not liable for any damages arising from its use. This package performs pattern-based filtering only and does not guarantee comprehensive content moderation. Users should implement additional verification layers for production use. See LICENSE for full terms.

---

Built by the Mofilo team