Ask Codex to turn a bank export into a tidy spreadsheet

How to do it

1. 1

   Download a CSV from the bank and make a copy. Remove the account number and counterparty names if you do not need them. Keep a few rows as a sample.

2. 2

   Write down the column names as they actually appear, and whether the bank uses a comma or a dot as the decimal mark. Do not assume the agent knows the format.

3. 3

   Ask Codex to write a script that reads the file and writes a new CSV with the columns you ask for. Amounts should be kept, not interpreted differently from what you described.

4. 4

   Read the script before you run it. It should only read the local file and write a new file, not send data anywhere.

5. 5

   Run the script on the copy and spot-check rows against the original. Run it on the rest only when the sample matches.

Example prompt

Write a Python script that reads bank.csv in the same folder. The first line is a header. The columns are named Dato, Tekst and Beløp. The decimal mark is a comma, and the thousands separator is a space.

Write ryddet.csv with the same three columns, amounts with a dot as the decimal mark, and no other columns. The script must not use the network. Do not guess other column names. If a row has the wrong number of fields, write the row number to stderr and skip it.