Definition: An instruction or information given to a model that specifies what it should do or how it should respond.
Simply put: A Prompt is what you tell the AI: “Do this for me.”
Examples:
- “Debug this code.”
- “Translate this text into Persian.”
- “Write tests for this API.”
Structure
A prompt usually has:
- Goal.
- Context.
- Constraints.
- Acceptance Criteria.
- Output Contract.
- Verification.
Example
Goal: Make the account deletion endpoint idempotent.
Context: The API is located in src/account; the project pattern follows the billing endpoints.
Constraint: The public API contract and the database schema must not change.
Acceptance Criteria: A repeated request should not return an error, and a regression test must be added.
Before editing, report the request path and the points of change.
- Note: More context isn’t always better. Use progressive disclosure. First the plan, then the related files, then necessary details.