Ship It Safely: Secrets, Data and Maintenance

Ship It Safely: Secrets, Data and Maintenance 1
⏱ 2 min readUpdated 13 October 2026

📘 Vibe Coding Course · Lesson 6 of 6

In this article
  1. Secrets
  2. Data
  3. A README your colleagues can follow
  4. Version it
  5. Plan for change

A tool that only you use can be rough. The moment colleagues depend on it, a few habits matter.

Advertisement

Secrets

  • Never paste API keys or passwords into code. Use environment variables or a .env file that is excluded from sharing.
  • Ask the assistant: “Move every key and password out of the code into environment variables and list them in the README.”
import os
API_KEY = os.environ["ANTHROPIC_API_KEY"]   # fails loudly if missing — good

Data

  • Keep real files out of the project folder; point the tool at them instead.
  • Don’t send personal or financial data to online AI tools unless your company allows it.
  • Write outputs to a new file; never overwrite the original.

A README your colleagues can follow

What it does: cleans the monthly bank export into statement_clean.xlsx
How to run:   python clean.py path/to/export.csv
Needs:        Python 3.12, pip install -r requirements.txt
Check:        the printed total must equal closing - opening balance
Owner:        your name, and who to call if it breaks

Version it

Ask the assistant to set up Git in the folder and commit after each working change. When a change breaks things, you can always go back to the last good version.

Plan for change

The bank will change its export format. A vendor will add a column. When that happens: drop the new file into sample, write what the output should be into expected, and ask the assistant to make the tests pass again. That is the whole maintenance process — and the reason we built the samples in Lesson 2.

You now have the complete loop: describe, build, verify, ship. Pick one chore from your week and build the tool for it.

✨ Ask AI about this article

Stuck on a step? Ask a question and the AI answers using this article.

Free · AI can be wrong