When should you use it?
- Every tracking tool you tried died because upkeep took too long
- You want answers grounded in your own numbers, not generic advice
- You keep repeating the same investing or spending mistakes
Not for: Handing an agent trading or payment access; Anyone not comfortable keeping account exports on their own machine
What do you need, and what do you get?
You need
- Account and portfolio exports
- A plain-language CLAUDE.md describing the assistant's role
- Optional: a data connector for public economic series
You get
- A folder the assistant keeps current with your figures
- A running log of decisions and the reasoning behind them
- Answers to questions about your financial health
How does it work, step by step?
- Write the role fileDescribe in CLAUDE.md what the assistant tracks, how it should talk to you and what it must never do, such as place orders.
- Load your own numbersDrop exports of accounts and holdings into the folder so every answer starts from your real situation.
- Add outside data if usefulConnect a source of public economic series only if you actually make choices based on it.
- Keep a decision diaryAsk the assistant to record each money call you make, with the reason, so later reviews can spot patterns.
- Review monthlyHave it update the figures from fresh exports and flag anything that drifted from your stated plan.
The prompt
Our own prompt, inspired by adlrocha and written to work in Claude Code and Claude. Values in lime are placeholders.
You are my personal finance assistant working in this folder. Read the account and portfolio exports here and keep a summary file current. When I ask a question, answer from my own figures first and say which file each number came from. Keep a decisions log: each time I tell you about a money choice, record the date, what I did and why. Never place orders, move funds or contact anyone. Once a month, compare my current position to the plan written at the top of this file and list anything that drifted.
Inspired by adlrocha: @adlrocha - Agents will make Code and Apps obsolete. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- A vague role file gives vague answers; spell out what counts as a good recommendation for you.
- Market data is context, not a signal to act on; keep final calls yours.
Our take
This is one of the clearest arguments for skipping the app entirely. The author tried the usual ladder and landed on instructions plus files he owns. What makes this work is that the role file is specific. It says what the assistant tracks, how it should speak to you and what it must never do, such as place orders, so each session starts from the same rules. Drop account exports into the folder so answers start from your real numbers rather than generic advice. The decision diary is the habit we would copy even without the rest; recording each money call with its reason lets a later review spot repeated mistakes. Add a public economic data source only if you actually make choices based on it, and treat market data as context, not a signal. Keep the exports on your own machine, encrypt them if you can, and never give the assistant payment or trading access.
Where the time goes back: It removes the maintenance chore that killed every earlier tracker, so the habit actually survives.
FAQ
Do I need to write code for this?
Why log decisions?
Source and credit
Inspired by adlrocha. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- @adlrocha - Agents will make Code and Apps obsolete
- Creator
- adlrocha
- Published on
- adlrocha.substack.com (Substack)
- Original date
- Feb 15, 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 24, 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.