Describe the input, expected output, and one realistic example before requesting code. Decide where the data lives and what happens when it is missing. Ask for acceptance checks based on the user's goal, including a failure case.
Run generated code in an appropriate development environment and inspect the result. Try a small copy of your data before using a script on the original files.
Start with one recurring annoyance and one useful screen. These prompts help you specify a small app, clean a messy dataset, or build a tool whose behavior you can inspect.
