When should you use it?
- Card statements show 'Amazon' but not what was bought
- You want a budget view without a paid app
- You will take ten minutes to verify numbers
Not for: Sharing statements you are not comfortable sending to an AI provider; Anything that needs bank login access
What do you need, and what do you get?
You need
- Exported bank and card transactions
- Order history export from your main online store
- A short instructions file naming folders and rules
You get
- A local dashboard of spending by category and month
- Order-level detail behind large retail charges
How does it work, step by step?
- Make a project folderPut statements in one subfolder and choose another for outputs.
- Write the house rulesNote in the instructions file where data sits, what currency, rounding and categories you want.
- Ask for the dashboardRequest a local page that groups spending by category and month and links retail charges to orders.
- Spot-checkPick several totals and compare them with the original statements before relying on anything.
- Use a second session as reviewerIf unsure, open a fresh session and ask it to audit the first one's calculations.
The prompt
Our own prompt, inspired by Michael Crist and written to work in Claude Code. Values in lime are placeholders.
In this folder, /data has my bank and card exports and my online order history. Build a local HTML dashboard of household spending by category and month. Match large retail card charges to the orders behind them and split them by item category. Remove duplicates across overlapping exports and exclude transfers between my own accounts. Round to whole dollars. Write the totals you used to a CSV so I can check them against the statements. Do not upload anything anywhere.
Inspired by Michael Crist: The Non-Technical Person's Complete Guide to Claude Code. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Duplicate rows across overlapping exports inflate totals; ask it to dedupe by date, amount and merchant.
- Refunds and transfers between your own accounts can look like spending.
Our take
Michael Crist insists on the step most people skip: checking the numbers. Any dashboard, whether built by an agent or a spreadsheet wizard, is only useful once you trust it, and comparing a handful of monthly totals and individual transactions with the original statements takes minutes. The clever addition is joining Amazon order history to card lines. A single large charge from a big retailer often hides groceries, gifts, household supplies and electronics, and splitting it by item makes categories meaningful. Working in a local folder rather than a chat window matters too, because next month you just drop new exports in and ask for a refresh, with the instructions file carrying the rules forward. Before starting, remove or mask full account numbers from the files, and keep the folder off shared drives. Ask for a short list of transactions it could not categorize instead of letting it guess silently. A second session reviewing the first one's categories, as he suggests, catches systematic errors like refunds counted as spending.
Where the time goes back: Categorizing months of transactions by hand is replaced by a short request and a quick verification pass.
FAQ
Do I need to code?
How do I trust the numbers?
Source and credit
Inspired by Michael Crist. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- The Non-Technical Person's Complete Guide to Claude Code
- Creator
- Michael Crist
- Published on
- michaelcrist.substack.com (Substack)
- Original date
- Mar 18, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Claude Code
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.