hosaypeng Extract Bank Statement
Extract transactions from bank account PDF statements and output pipeline-compatible CSVs.
Trigger
Use when the user says:
- "extract transactions from [PDF]"
- "convert bank statement PDF to CSV"
- "parse this bank statement"
- "extract data from bank statement"
Supported Banks
| Bank | Statement type | Script |
|---|---|---|
| Trust Bank | Savings account (monthly) | extract_trust_bank_savings.py |
| Trust Bank | Credit card (monthly) | extract_trust_bank_cc.py |
Usage
Savings account — single file
python3 ~/.claude/skills/extract-bank-statement/scripts/extract_trust_bank_savings.py \
/path/to/statement.pdf \
/path/to/output.csv
Credit card — single file
python3 ~/.claude/skills/extract-bank-statement/scripts/extract_trust_bank_cc.py \
/path/to/statement.pdf \
/path/to/output.csv
Omit the output path on either script to print to stdout (useful for preview).
Batch: all PDFs in a directory
SCRIPT=~/.claude/skills/extract-bank-statement/scripts/extract_trust_bank_cc.py
for f in /path/to/statements/pdf/*.pdf; do
name=$(basename "$f" .pdf)
python3 "$SCRIPT" "$f" "/path/to/output/${name}.csv"
done
Output Schema
Savings account outputs 7 columns matching the finance pipeline schema (date, account, description, amount, type, category, balance).
Credit card outputs 9 columns — the 7 pipeline columns plus 2 supplementary FCY columns:
| Column | Format | Example | Notes |
|---|---|---|---|
| date | YYYY-MM-DD | 2025-07-21 | |
| account | "Trust Bank" | Trust Bank | |
| description | as-is from statement | DE LUNA HOTEL KUALA LUMPUR MY | |
| amount | positive float | 39.54 | always SGD |
| type | "credit" or "debit" | debit | |
| category | empty (pipeline categorizes) | ||
| balance | empty (not in statement) | ||
| fcy_amount | positive float or empty | 130.00 | CC only; original foreign currency amount |
| fcy_currency | ISO 4217 code or empty | MYR | CC only; pipeline ignores, useful for analysis |
Savings Account Rules
- Credits: SGD amount prefixed with
+in the source PDF - Debits: SGD amount with no prefix
- Skipped rows: "Previous balance", "Closing balance"
- Empty statements (all zeros): script prints a warning, no CSV rows emitted
- Year is derived from the "Statement period" header on page 1
Credit Card Rules
- Credits: SGD amount prefixed with
+(cashback, repayments) - Debits: SGD amount with no prefix
- FCY transactions: exchange-rate and FCY-amount lines are skipped; only SGD amount is captured
- Skipped rows: "Previous balance", "Total outstanding balance"
- Section headers skipped: "Cashback credit card", "Link credit card", "Loans, Fees, Charges and Repayments"
- Year is derived from the "Statement cycle" header on page 1; cross-year cycles (e.g. Dec–Jan) are handled correctly
- Two PDF layouts supported automatically:
- V1 (Feb–Aug 2025): single "Posting date" column
- V2 (Sep 2025+): "Transaction date" + "Posting date" columns; transaction date is used
Dependency
Requires pymupdf:
pip install pymupdf
Workflow Integration
After extracting CSVs, place them in the appropriate source directory
(e.g., trust_bank_account_statements/) and run the finance pipeline using the
pipeline_script path from ~/.claude/skills/extract-bank-statement/config.json.
The pipeline will merge, categorize, and validate them into the master CSVs.