Generate Unit Tests for a Function

This prompt asks a model acting as a TDD expert to write a full test suite for one function: happy path, edge cases, and failure modes, with sentence-style test names and mocked dependencies. Reach for it when you have working code with thin or no coverage and want a thorough starting suite fast. The explicit edge-case checklist (empty, boundary, null/undefined) is what separates this from the one-or-two-tests an LLM writes by default.

What you can expect back

import { describe, it, expect } from 'vitest';
import { calculateDiscount } from './calculateDiscount';

describe('calculateDiscount', () => {
  it('applies no discount for the free tier', () => {
    expect(calculateDiscount(100, 'free')).toBe(100);
  });

  it('applies a 10 percent discount for the pro tier', () => {
    expect(calculateDiscount(100, 'pro')).toBe(90);
  });

  it('applies a 20 percent discount for the enterprise tier', () => {
    expect(calculateDiscount(50, 'enterprise')).toBe(40);
  });

  it('returns zero when the price is zero', () => {
    expect(calculateDiscount(0, 'pro')).toBe(0);
  });

  it('throws when the price is negative', () => {
    expect(() => calculateDiscount(-1, 'free')).toThrow('price must be non-negative');
  });
});

Illustrative example — your results will vary by tool and inputs.

Get sharper results

• 01Name the exact framework version — Jest, Vitest, and pytest differ in mocking syntax, and an unspecified framework gives you tests you have to rewrite.
• 02Paste the function's real dependencies or their signatures so the mocks match your actual interfaces rather than invented stand-ins.
• 03Ask it to flag any branches it could not test because of how the code is written — that often reveals a function that needs refactoring to be testable.
• 04If the function is pure, tell it so; the model can then skip mocking entirely and focus the budget on boundary and property-style cases.
• 05Request one parameterized/table-driven test for the tier-style variations to keep the suite compact instead of one near-duplicate test per case.

Adapt it for your case