Plan a Feature Before Coding
Coding agents do better work when they plan first. This prompt asks for a short plan you can approve or correct before a single file changes.
Works with Claude, ChatGPT and Gemini3 fields
Before writing any code, make a plan for this feature: Let users export their data as CSV. Constraints: no new dependencies; must work for 100k rows; keep the API backward compatible Done means: a button on the settings page downloads a CSV; tests pass The plan should list: 1. Files you will create or change, and why. 2. Steps in order, each small enough to review. 3. Risks and how you will test them. 4. Questions you need me to answer first. Stop after the plan and wait for my go-ahead.
Everything you type stays in your browser. Remove passwords, keys and personal data before pasting anything into an AI tool.
Why this prompt is written this way
- Plan, then stop. Waiting for approval lets you fix a wrong approach before it costs time.
- Definition of done. A clear finish line keeps the agent from over-building.
- Questions up front. Unknowns surface before the agent guesses its way around them.
Example answer
Illustrative and shortened, to show the shape of a good answer. Real answers vary by model.
Files: api/export.py (new), web/Settings.tsx (button), tests/test_export.py. Steps: 1) streaming CSV writer 2) endpoint 3) button 4) tests. Risk: memory with 100k rows, so stream instead of building the file in memory. Question: which columns should the export include?
Tips
- In Claude Code you can also use plan mode for the same effect.
- Answer the questions, then reply 'go ahead with steps 1-2' to keep changes small.
How to use it
- Replace the example values with your own. The prompt on the right updates as you type.
- Press Copy prompt and paste it into any AI chat, or open it straight in Claude or ChatGPT.
- Save it as prompt.md if you want to reuse it.