| name | chitti-upi |
| description | Chitti UPI Fraud Guard — Indian-family-friendly fraud-risk classifier for UPI / digital payments. User pastes a payment request, SMS, WhatsApp message, merchant name, or what a caller told them; Chitti returns HIGH (red) / MEDIUM (orange) / LOW (green) risk plus a one-line spoken warning and the next safe step. Educational RBI 2026 rule cards (2FA · 1-hour cooling lag · Trusted Person · Kill Switch). Always appends "Fraud hone par turant 1930 pe call karo ya cybercrime.gov.in pe report karo." and "Chitti ek AI warning tool hai — yeh payment block nahi kar sakta." Use on chitti_upi.html or for any UPI fraud / scam SMS / suspicious caller question. |
Chitti UPI Fraud Guard — top-level skill
This skill describes the Chitti UPI Fraud Guard product as a whole.
It loads when the user asks about UPI fraud, suspicious payment
requests, KYC / lottery / refund / electricity-disconnection scams, or
the RBI 2026 framework (2FA · cooling lag · Trusted Person · Kill
Switch).
Repo layout
chitti-upi/
├── frontend/index.html mirror of workspace-root chitti_upi.html
└── backend/
├── main.py Flask app · CORS · /api/upi/* registration
├── config.py settings (DEEPSEEK_API_KEY · model · CORS · token caps)
├── requirements.txt flask · flask-cors · gunicorn · httpx
├── runtime.txt Python 3.11
├── render.yaml Railway Blueprint
├── services/
│ └── upi_service.py DeepSeek wrapper + CHITTI_UPI_FRAUD_PROMPT + JSON-mode parsing + RBI 2026 cards
└── routes/
└── upi.py Blueprint /api/upi/*
Endpoint surface (Flask Blueprint, prefix /api/upi)
| Method | Path | Purpose |
|---|
| POST | /check | Body {text, language?} → {risk, reason, warning, indicators[], actions[], legal_lines[]} |
| GET | /rules | Static RBI 2026 educational cards (icon · title · body · speak text) |
| GET | /health | DeepSeek key configured? |
risk is always one of HIGH (red), MEDIUM (orange), LOW (green).
The frontend MUST render the colour band and read warning aloud.
legal_lines is always two strings — the frontend appends both to
the visible warning AND speaks them after the warning.
Consent gating
localStorage.chitti_upi_consent_given gates every feature. The T&C
modal has 6 sections (mirrors Vaani structure) with 🔊 buttons:
- What Chitti UPI Guard is (and is not — it does NOT block payments)
- Legal Compliance (RBI Master Directions on Digital Payments · DPDP Act 2023)
- Permissions Required (none — everything is text-pasted by the user)
- Data Privacy (no payment data, no UPI handle stored)
- Emergency and Liability (cyber-fraud → 1930 / cybercrime.gov.in)
- Grievance Redressal (sire@sahayai.in)
Cross-product hooks
- UPI Guard → Vaani: when a HIGH verdict is shown, the page
surfaces a "🚨 Talk to Chitti Vaani for SOS" deep-link.
- Scanner → UPI Guard: when Scanner detects an insurance /
payment-related document, it can deep-link here for a premium-safety
check.
Live URLs