SKILL: Finance & Expense Tracking

1. Overview

Manage Pratik's financial life by monitoring Gmail for transaction alerts, maintaining a monthly Markdown ledger, and reconciling credit card statements.

Work Directory: /root/clawd/finance Card Registry: /root/clawd/finance/cards.md

2. Trigger

• Source: Gmail (via gog skill).
• Query: label:Finance -label:Processed (Search for unread transaction alerts).
• Frequency: Every 1 hour (via Heartbeat or Cron).

3. The Ledger (File Structure)

• Path: /root/clawd/finance/YYYY/YYYY-MM.md (e.g., 2026/2026-02.md).
• Template:

  # February 2026 - Expenses
  
  ## 🚨 Pending Clarification
  - [ ] **₹4,500** at "M/S RANDOM" on Feb 02 - *User asked via WA*
  
  ## 💳 Transaction Log
  | Date | Merchant | Amount | Category | Card (Last 4) | Notes |
  | :--- | :--- | :--- | :--- | :--- | :--- |
  

4. Processing Rules

A. Parsing Email Alerts

1. Extract: Date, Merchant, Amount, Currency, Card Last 4 Digits.
2. Currency:
   • If INR (₹): Log directly.
   • If Foreign ($/€): Perform web_search for current rate. Log converted INR in Amount column. Add original currency to Notes.
3. Type:
   • Debit/Spend: Positive amount.
   • Credit/Refund: Negative amount (e.g., -₹500.00). Check for keywords: "Credited", "Refund", "Reversal".

B. Validation & Logic

1. Duplicate Check: Before appending, read the last 20 lines of the ledger. If Date + Merchant + Amount matches -> STOP. Send WhatsApp: "Duplicate detected: [Details]. Real or Glitch?"
2. Clarification: If Merchant is opaque (e.g., "RAZORPAY*"), append to ## 🚨 Pending Clarification and WhatsApp user.

C. Statement Parsing (PDFs)

1. Trigger: Email with attachment *.pdf from a known Bank (check cards.md).
2. Password Cracking:
   • Read cards.md for PII (Name, DOB, PAN, Mobile).
   • Try common formats:
     • PRAT10062000 (First 4 Name + DOB)
     • 10062000 (DOB)
     • 1006XXXX (DDMM + Last 4)
     • XXXX1006 (Last 4 + DDMM)
   • Action: If extraction fails, WhatsApp user: "Cannot open [Bank] statement. What is the password format?"
   • Action: If successful, extract table rows and reconcile against the Markdown ledger.

5. Reporting

• Weekly Summary: On Monday morning, provide a quick total burn rate for the month.
• Commands: If user asks "How much spent on Food?", use Python to parse the Markdown table and sum by category.

6. Maintenance

• Always apply Gmail label Processed after successful logging to prevent loops.