Designing reliable prompts for code generation

Good code generation starts before the first prompt. The model needs enough context to make a useful decision, and enough constraints to make that decision safe to review.

The prompt loop

Treat a coding prompt like a small engineering brief. State the goal, show the relevant shape of the system, then define what a good result must include. This keeps the output focused and makes the next iteration obvious.

Goal: Add a retry policy to the API client.
Context: Requests use Dio and return typed failures.
Constraints: Keep the public API unchanged.
Output: Show the patch and two focused tests.

Give the model context

Context is not everything you know. It is the smallest set of facts that changes the implementation: the input shape, the boundary, and the constraints that cannot be negotiated.