Unit Test Generator Prompt: Meaningful Tests with Claude

Use this prompt when you need tests for existing code or want to add coverage before a refactor. It asks Claude to plan the cases first, follow your project’s conventions, and report possible bugs instead of writing tests that lock in wrong behavior.

The prompt

You are a senior engineer who writes clear, behavior-focused unit tests. The tests will be maintained by the team, so readability matters as much as coverage.

<code_under_test>
{{CODE}}
</code_under_test>

<context>
Language and test framework: {{FRAMEWORK}}
Example of an existing test in this project (for style): {{EXAMPLE_TEST_OR_NONE}}
How to run the tests: {{TEST_COMMAND}}
</context>

<instructions>
1. Summarize what the code is supposed to do, based on names, docs and logic. Mark inferences as "Assumption".
2. Before writing code, list the test cases in a table: happy paths, boundaries (empty, zero, max, unicode), invalid input, error paths, and state or ordering effects.
3. Write tests that check observable behavior through the public interface. Mock only external boundaries (network, clock, filesystem, randomness).
4. Use one clear assertion focus per test and descriptive test names.
5. Match the style of the example test when one is provided.
6. If the code looks buggy for a case, write the test for the correct behavior, mark it clearly, and explain the suspected bug. Do not change expectations to fit the current output.
7. If you can run commands, run the tests with the given command and report the results.
</instructions>

<output_format>
## Behavior summary
## Test plan
| Case | Input | Expected | Type |
## Tests
One code block with the complete test file.
## Suspected bugs
## Not covered and why
</output_format>

How to use it

• claude.ai: paste the code and one existing test for style.
• Claude Code: the best fit, because Claude can read the module, find existing tests, and run the suite. Save it as a skill at .claude/skills/write-tests/SKILL.md and pass the file path with $ARGUMENTS; see the skills guide. For a hard gate, a Stop hook can run your test command before Claude finishes; see the hooks guide.
• API: send the code first and the instructions after it.

Variables

Variable	What to put there
{{CODE}}	Function, class or module under test
{{FRAMEWORK}}	“TypeScript + Vitest”, “Python + pytest”, “Go testing”
{{EXAMPLE_TEST_OR_NONE}}	One existing test file, or “none”
{{TEST_COMMAND}}	npm test, pytest -q, go test ./...

Example output (illustrative)

## Test plan
| Case | Input | Expected | Type |
| empty cart | [] | total 0 | boundary |
| 100% discount | price 50, discount 1.0 | total 0, no negative | boundary |
| negative quantity | qty -1 | throws ValidationError | invalid input |

## Suspected bugs
- A discount above 1.0 produces a negative total; no validation exists.

Tips and variations

• Characterization tests: before a refactor, ask Claude to test current behavior exactly, bugs included, and label them. Then use the refactor plan prompt.
• Property-based tests: ask for properties (for example “sorting is idempotent”) using your language’s property-testing library.