Hadith Verifier

AI-powered tool to detect fabricated and weak hadiths spreading on social media.

Live: https://hadithverifier.com
Vercel: https://hadith-verifier-vp57.vercel.app
GitHub: https://github.com/Farhod75/hadith-verifier

Supports Facebook, Instagram, WhatsApp, and Telegram posts in Uzbek, Arabic, Russian, English, and Tajik.

Features

Multi-language — analyzes posts in any language, generates replies in EN / UZ / AR / RU / TJ
Screenshot analysis — upload Facebook/Instagram screenshots for Claude Vision OCR analysis
Dua corrector — checks Arabic dua text and transliteration for errors across 4 scripts
3-tier source system — Dorar.net, Sunnah.com, IslamQA, HadeethEnc, IslamWeb
Red flags detection — identifies specific fabrication patterns (chain messages, reward promises, etc.)
Ready-to-paste comments — compassionate correction reply with direct links to authenticated sources
Admin queue — persistent Supabase storage, human review workflow
Telegram bot — users forward suspicious posts, bot replies instantly (full automation)
Security audit — prompt injection resistance, output quality validation on every deploy
Auto-Fix Agent — Python agent monitors CI failures and auto-generates fixes via Claude API

Core principle: AI flags, humans decide. No auto-delete or auto-ban.

Architecture

┌─────────────┐    ┌──────────────────┐    ┌─────────────────────────┐    ┌────────────────────┐
│    USER     │    │  Next.js 14 UI   │    │   /api/analyze route    │    │  Anthropic Claude  │
│             │───▶│  hadithverifier  │───▶│  ① sanitizeInput()      │───▶│  claude-sonnet-4-6 │
│ Pastes post │    │  .com            │    │  ② langInstruction      │    │  Vision + Analysis │
│ or uploads  │    │  4 tabs:         │    │  ③ JSON extraction      │◀───│  JSON response     │
│ screenshot  │◀───│  Analyze / Dua / │◀───│  ④ validateOutput()     │    └────────────────────┘
│             │    │  Sources / Admin │    │  ⑤ calculateSeverity()  │
└─────────────┘    └──────────────────┘    └────────┬────────────────┘
                                                    │
                                    ┌───────────────┼───────────────┐
                                    ▼               ▼               ▼
                           ┌──────────────┐ ┌───────────┐ ┌───────────────┐
                           │   Supabase   │ │   Slack   │ │   Telegram    │
                           │ flagged_posts│ │  Webhook  │ │   Bot alerts  │
                           │ admin review │ │ CRITICAL/ │ │   Railway     │
                           │ workflow     │ │ HIGH only │ │   deployment  │
                           └──────────────┘ └───────────┘ └───────────────┘

Source authority tiers

Tier	Sources	Why trusted
1 (Highest)	Dorar.net, Sunnah.com, HadeethEnc.com	Full 9 canonical collections, classical scholar grading
2 (High)	IslamQA.info, IslamWeb.net, Yaqeen Institute, Islamhouse.com	Peer-reviewed fatwa databases, multilingual
3 (Standard)	HadithAPI.com, AboutIslam.net, AlSunnah.com	REST APIs and Q&A portals

Severity scoring

Verdict	Confidence	Severity	Action
fabricated	high	🔴 CRITICAL	Save to Supabase + Slack alert
fabricated	medium/low	🟠 HIGH	Save to Supabase + Slack alert
weak	high	🟠 HIGH	Save to Supabase + Slack alert
weak	medium/low	🟡 MEDIUM	Save to Supabase, no alert
unclear	any	🟡 MEDIUM	Save to Supabase, no alert
authentic	any	🟢 LOW	Not saved, no alert
no_hadith	any	🟢 LOW	Not saved, no alert

Quick start (local)

1. Clone and install

git clone https://github.com/Farhod75/hadith-verifier
cd hadith-verifier
npm install

2. Set up environment variables

cp .env.example .env.local
# Edit .env.local and add your keys

Get your keys from:

Anthropic API key — https://console.anthropic.com (separate from claude.ai subscription)
Supabase — https://supabase.com (free account)

3. Set up Supabase database

Create a free project at supabase.com
Go to SQL Editor → New Query
Run the schema from supabase_schema.sql in this repo, or paste:

CREATE TABLE flagged_posts (
  id UUID DEFAULT gen_random_uuid() PRIMARY KEY,
  post_text TEXT NOT NULL,
  verdict TEXT NOT NULL,
  confidence TEXT,
  severity TEXT,
  claim_summary TEXT,
  analysis TEXT,
  suggested_comment TEXT,
  lang TEXT DEFAULT 'en',
  sources JSONB,
  red_flags JSONB,
  reviewed BOOLEAN DEFAULT false,
  created_at TIMESTAMPTZ DEFAULT now()
);

ALTER TABLE flagged_posts DISABLE ROW LEVEL SECURITY;
GRANT ALL ON flagged_posts TO service_role;
GRANT ALL ON flagged_posts TO anon;

Copy your project URL and keys to .env.local

4. Run the web app

npm run dev
# Open http://localhost:3000

5. Run the Telegram bot (optional)

pip install -r requirements.txt
# Add TELEGRAM_BOT_TOKEN to .env.local
# Create bot at @BotFather on Telegram first
python telegram_bot.py

API

POST /api/analyze

Accepts JSON (text) or multipart/form-data (image upload):

{
  "postText": "string",
  "lang": "en | uz | ar | ru | tj"
}

Or with image:

Content-Type: multipart/form-data
image: <PNG or JPG file>
lang: en

Response:

{
  "verdict": "fabricated|weak|authentic|unclear|no_hadith",
  "confidence": "high|medium|low",
  "severity": "CRITICAL|HIGH|MEDIUM|LOW",
  "claim_summary": "string",
  "analysis": "string",
  "suggested_comment": "string",
  "red_flags": ["string"],
  "references": [
    { "source": "string", "url": "string", "authority": "tier1|tier2|tier3" }
  ],
  "extracted_text": "string (image path only)"
}

GET /api/queue

Returns all flagged posts ordered by created_at DESC, limit 50.

PATCH /api/queue

Marks a post as reviewed. Body: { "id": "uuid" }

DELETE /api/queue

Deletes a post. Body: { "id": "uuid" }

POST /api/dua

Checks Arabic dua for errors and returns corrections + transliterations.

Security architecture

Layer	What it does
`sanitizeInput()`	Strips HTML/script tags, detects 6 prompt injection patterns, rejects input > 5000 chars
`validateOutput()`	Enforces verdict/confidence/severity enums, filters URLs to trusted domains whitelist
JSON extraction	Uses `indexOf('{')` + `lastIndexOf('}')` — handles Cyrillic preamble and Unicode artifacts
Supabase RLS	Disabled — server routes use `SUPABASE_SERVICE_ROLE_KEY`, not anon key
Vercel env vars	`ANTHROPIC_API_KEY` stored as Vercel secret, never exposed to client

What the app can and cannot do

Action	Possible
Analyze any post text or screenshot	✅ Yes
Generate correction comment in 5 languages	✅ Yes
Post comment automatically on Facebook/Instagram	❌ No — Meta API restricts this
Delete post in your own Facebook group	✅ Yes — via group admin tools manually
Telegram bot auto-reply	✅ Yes — full automation
Report to Facebook/Instagram	⚠️ Manual only (open their report flow)

CI/CD pipeline

GitHub Actions runs on every push to main. Tests run against production URL (no local build needed).

# .github/workflows/ci.yml
timeout-minutes: 55
BASE_URL: https://hadithverifier.com
--workers=1  # prevents Claude API rate limiting
--project=chromium  # cost-efficient, Firefox/WebKit run locally

Step	Spec file	What it covers
API tests	`tests/api.spec.ts`	Language output (UZ/AR/RU/EN/TJ), response structure, hallucination detection, image upload path
E2E tests	`tests/hadith-verifier.spec.ts`	Full UI flow, language switching, copy button, result rendering
Audit tests	`tests/audit.spec.ts`	Prompt injection resistance, output quality, Islamic greeting compliance, source integrity, content safety
Severity tests	`tests/severity.spec.ts`	Severity scoring matrix validation
Accessibility	`tests/accessibility.spec.ts`	WCAG compliance checks

# Run full test suite locally
npx playwright test --project=chromium

# Run specific suite
npx playwright test tests/audit.spec.ts --project=chromium
npx playwright test tests/api.spec.ts --project=chromium

# Run against production
BASE_URL=https://hadithverifier.com npx playwright test --project=chromium

# Python tests
cd tests/python && pip install -r requirements.txt
pytest test_analyze_api.py -v

Auto-Fix Agent

agents/playwright_agent.py monitors CI failures and automatically fixes them:

Trigger — workflow_run event fires when CI completes with failures
Fetch — reads GitHub Actions check-run annotations for failed tests
Load — reads agents/knowledge/fix_patterns.md (35 documented patterns)
Ask Claude — sends failure context to claude-sonnet-4-6 for fix generation
Apply — commits fix to a new branch
PR — opens GitHub Pull Request automatically

Requires: Settings → Actions → General → "Read and write permissions" + "Allow GitHub Actions to create PRs"

Deploy to Vercel (web app)

Connect your GitHub repo to Vercel dashboard and add these environment variables:

ANTHROPIC_API_KEY
NEXT_PUBLIC_SUPABASE_URL
NEXT_PUBLIC_SUPABASE_ANON_KEY
SUPABASE_SERVICE_ROLE_KEY
SLACK_WEBHOOK_URL
TELEGRAM_ALERT_BOT_TOKEN
TELEGRAM_ALERT_CHAT_ID

Then:

vercel --prod --force

Deploy Telegram bot to Railway (free)

Go to railway.app → New Project → Deploy from GitHub
Select your repo
Add environment variables: ANTHROPIC_API_KEY, TELEGRAM_BOT_TOKEN
Set start command: python telegram_bot.py
Deploy — bot runs 24/7 on Railway's free starter plan

How to create your Telegram bot

Open Telegram → search @BotFather
Send /newbot → choose name: Hadith Verifier
Copy the token → add to .env.local as TELEGRAM_BOT_TOKEN
Send /setcommands and paste:

start - Welcome message
help - How to use
lang_en - Reply in English
lang_uz - Reply in Uzbek
lang_ar - Reply in Arabic
lang_ru - Reply in Russian

Tech stack

Layer	Technology
Frontend	Next.js 14 + TypeScript + Tailwind CSS
AI Engine	Anthropic Claude API (`claude-sonnet-4-6`)
Database	Supabase (PostgreSQL)
Deployment	Vercel (web app) + Railway (Telegram bot)
Testing	Playwright (TypeScript) + pytest (Python)
CI/CD	GitHub Actions
Bot	Python + Telegram Bot API

Known issues

text-gray-400 hint text has color contrast ratio of 2.53:1 (below 4.5:1 WCAG 2.1 AA). Fix: replace with text-gray-700.
Image upload parse error may occur on very large screenshots — mitigated by max_tokens: 3000 on image path.

Project structure

hadith-verifier/
├── app/
│   ├── api/
│   │   ├── analyze/route.ts    # Main hadith analysis API
│   │   ├── dua/route.ts        # Dua correction API
│   │   └── queue/route.ts      # Admin queue API
│   └── page.tsx                # Main UI (4 tabs)
├── agents/
│   ├── playwright_agent.py     # Auto-Fix Agent
│   └── knowledge/
│       └── fix_patterns.md     # 35 documented fix patterns
├── tests/
│   ├── api.spec.ts             # API tests
│   ├── hadith-verifier.spec.ts # E2E UI tests
│   ├── audit.spec.ts           # Security audit tests
│   ├── severity.spec.ts        # Severity scoring tests
│   ├── accessibility.spec.ts   # Accessibility tests
│   ├── fixtures/
│   │   └── test-data.ts        # Test fixtures
│   └── python/
│       └── test_analyze_api.py # Python pytest suite
├── .github/workflows/
│   ├── ci.yml                  # Main CI/CD pipeline
│   └── auto-fix.yml            # Auto-Fix Agent trigger
├── telegram_bot.py             # Telegram bot
├── supabase_schema.sql         # Database schema
└── CLAUDE.md                   # Claude Code context

Author

Farhod Elbekov — github.com/Farhod75

Built as a portfolio project demonstrating Claude API integration, full-stack TypeScript, Python automation, AI-powered QA workflows, and real-world Islamic content moderation tooling.

Certifications: ISTQB CT-AI (March 2026) · ISTQB CTFL v4.0 (Feb 2026) · CT-GenAI (In Progress)

Name		Name	Last commit message	Last commit date
Latest commit History 105 Commits
.github/workflows		.github/workflows
agents		agents
app		app
lib		lib
playwright-report		playwright-report
test-results		test-results
tests		tests
.env.example		.env.example
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
QA_STANDARDS.md		QA_STANDARDS.md
README.md		README.md
next-env.d.ts		next-env.d.ts
next.config.js		next.config.js
package-lock.json		package-lock.json
package.json		package.json
playwright.config.ts		playwright.config.ts
postcss.config.js		postcss.config.js
requirements.txt		requirements.txt
supabase_schema.sql		supabase_schema.sql
tailwind.config.js		tailwind.config.js
telegram_bot.py		telegram_bot.py
tsconfig.json		tsconfig.json
tsconfig.tsbuildinfo		tsconfig.tsbuildinfo

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Hadith Verifier

Features

Architecture

Source authority tiers

Severity scoring

Quick start (local)

1. Clone and install

2. Set up environment variables

3. Set up Supabase database

4. Run the web app

5. Run the Telegram bot (optional)

API

POST /api/analyze

GET /api/queue

PATCH /api/queue

DELETE /api/queue

POST /api/dua

Security architecture

What the app can and cannot do

CI/CD pipeline

Auto-Fix Agent

Deploy to Vercel (web app)

Deploy Telegram bot to Railway (free)

How to create your Telegram bot

Tech stack

Known issues

Project structure

Author

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

Hadith Verifier

Features

Architecture

Source authority tiers

Severity scoring

Quick start (local)

1. Clone and install

2. Set up environment variables

3. Set up Supabase database

4. Run the web app

5. Run the Telegram bot (optional)

API

POST /api/analyze

GET /api/queue

PATCH /api/queue

DELETE /api/queue

POST /api/dua

Security architecture

What the app can and cannot do

CI/CD pipeline

Auto-Fix Agent

Deploy to Vercel (web app)

Deploy Telegram bot to Railway (free)

How to create your Telegram bot

Tech stack

Known issues

Project structure

Author

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages