You get a runnable Python script, not a sketch. These five Claude prompts turn a bare idea — scrape a page, rename a folder of files, call an API — into code with the imports, error handling, and a quick test already in place. Copy one, fill in the brackets, and paste it straight into Claude.
What to know before you paste
- On a newer Claude model you can ask for the whole script at once. On a smaller or faster tier, split the work so one prompt asks for the plan and the next asks for the code.
- Claude reads long pastes well, so include your existing script and the exact error text instead of a summary of it.
- Name your Python version and the libraries you will allow. Standard library keeps scripts portable; add packages only when you need them.
The prompts
1. Script from a plain-English spec
You are a Python developer. Write a standalone script that does the following: [task in plain English]. It reads input from [input source] and writes results to [output location]. Use only the standard library plus [allowed libraries]. Add type hints, a module docstring, and a __main__ guard. Print a one-line summary when it finishes.What it does: Turns a vague request into a complete, runnable script with sane defaults.
How to use: replace [task in plain English], [input source], [output location], and [allowed libraries]; keep the required structure and guard as is.
2. Add error handling and logging
Here is my Python script: [paste script]. Rewrite it to fail gracefully. Wrap file and network calls in try/except, log with the logging module at INFO for normal runs and ERROR for failures, and exit with a non-zero status when [critical step] fails. Explain each change in one sentence after the code.What it does: Makes a fragile script survive bad input, missing files, and dropped connections.
How to use: replace [paste script] with your current file and [critical step] with the step that must not fail silently.
3. Refactor into small functions
Take this script and refactor it into small, testable functions: [paste script]. Each function should do one thing and take clear arguments. Keep the same output, but make [specific block] reusable so I can call it on its own. Show the refactored file, then list the new function signatures.What it does: Splits one long block into functions you can reuse and test separately.
How to use: replace [paste script] and [specific block]; if a function name matters to you, add it to the request.
4. Write pytest tests for a function
You are writing tests for this Python function: [paste function]. Write pytest tests covering the happy path, empty input, and a [edge case type] case. Use parametrize where it keeps the file short. Include the exact command to run the tests and the file name I should save them to.What it does: Produces a starter test file plus the command to run it.
How to use: replace [paste function] and [edge case type] (for example, “duplicate keys” or “negative amounts”).
5. Explain and document the script
Read this Python script and explain what it does to [audience, e.g. a new teammate]. Then add a module-level docstring, docstrings above every function, and inline comments only where the logic is non-obvious. Return the documented file plus a short usage example I can paste into a README.What it does: Turns code into something a teammate can read and run without asking you.
How to use: replace [audience] with who will read it; paste the script after the prompt.
FAQ
Can Claude write a Python script from scratch?
Yes, if you describe the inputs, outputs, and allowed tools. Vague requests get vague code, so spell out what the script reads and writes.
What if the generated script has a bug?
Paste the full traceback and the line it points to, then ask for the smallest fix. The exact error text usually names the cause.
Should I paste my whole script or just one part?
Paste the whole file for a refactor, since Claude needs to see how the pieces talk. Paste one function to have it explained or tested.
Can these prompts handle pandas, requests, or other packages?
Yes. List the packages in [allowed libraries] and add any version constraints. If a package is unusual, paste a short usage example so the model matches your API.
How do I keep the style consistent across scripts?
Reuse the same short style note at the end of every prompt: formatter, line length, and naming. Consistency comes from repeating that note yourself.
Want more like this? Browse the full free library at GuPrompt.