| name | java |
| description | Always load this skill when writing, modifying, creating, or moving Java or Kotlin source code, or when project setup has already chosen Java/Kotlin as the implementation language.
|
| version | 0.3.0 |
Java & Kotlin Conventions
JVM language coding conventions complementing the TDD and Clean Code skills.
We use modern language features to reduce boiler plate and get more to the point.
Java Version
When you need to know the latest Java version (e.g. for new projects, upgrades, or Dockerfiles),
query the Adoptium API:
https://api.adoptium.net/v3/info/available_releases
Use most_recent_feature_release (latest GA) by default.
Use most_recent_lts only when explicitly asked for LTS.
Code Style
Logic Expressions
- Try to prevent negating expressions, e.g., instead of
a != b ? doX() : doY()
use a == b ? doY() : doX().
Line Breaks
- Annotations can often go in the same line as the field or method they apply to.
- Java:
@Override void foo() { or @Test void should() {
- Kotlin:
override fun foo() { or @Test fun should() {
Local Type Inference
Java: Use var for local variable type inference where the type is clear from context.
var users = userRepository.findAll();
var count = items.size();
Kotlin: Type inference is the default. Prefer val (immutable) over var (mutable);
only use var when the variable must be reassigned.
val users = userRepository.findAll()
var count = items.size
Static Imports (Java) / Imports (Kotlin)
Java: Prefer static imports when they do not reduce readability.
This includes constants like MediaType.APPLICATION_JSON.
The context is most often sufficient to understand what it is, e.g., @Produces(APPLICATION_JSON).
Exception: do not statically import List.of(...), Map.of(...), or the like,
as a method name like of doesn't say, what it does.
The number of usages of the import is not relevant!
import static jakarta.ws.rs.core.MediaType.APPLICATION_JSON;
import static org.assertj.core.api.BDDAssertions.then;
import static org.mockito.BDDMockito.given;
Kotlin: There are no static imports. Import top-level functions and object/companion members directly.
The same readability rule applies — import short, context-sufficient names; avoid importing bare of.
import jakarta.ws.rs.core.MediaType.APPLICATION_JSON
import org.assertj.core.api.BDDAssertions.then
import org.mockito.kotlin.given
Javadoc / KDoc
IMPORTANT Use Markdown for doc comments to improve in-IDE readability.
Java uses Javadoc (/** */); Kotlin uses KDoc (same /** */ syntax, same Markdown support).
Comments
Add comments only if they add real value. Prefer self-explanatory code.
Comments should explain "why", not "what".
Exception Handling
Java: Hide checked exceptions within a method by wrapping them in RuntimeException and add a helpful message:
try{
return objectMapper.writeValueAsString(value);
} catch(JsonProcessingException e) {
throw new RuntimeException("could not write value as string",e);
}
Kotlin: Has no checked exceptions, so no wrapping is needed. When calling Java APIs that declare
checked exceptions, Kotlin treats them as unchecked — handle them only if recovery makes sense.
Log Output
System.out output is only acceptable for "normal" output in CLI tools.
System.err output is only acceptable for warnings and diagnostics (logs) in CLI tools.
In all other cases, use some logging API (preferably slf4j) and if necessary a library (preferably logback).
Constructors and Dependency Injection
When available, use the capabilities of a Dependency Injection framework like CDI.
Even though constructor injection is not commonly used in CDI,
it makes, e.g., unit-testing (when DI is not available) easier than field injection does.
Injecting with setters is very seldom a sensible option to choose.
If the value of a field doesn't depend on constructor parameters, then use a Field initializer:
- Java:
private final HttpClient httpClient = HttpClient.newHttpClient();
- Kotlin:
val httpClient = HttpClient.newHttpClient()
Visibility
Only well-defined APIs are public. Internals are not, but have limited visibility,
i.e. private if possible. If wider than private is needed:
- Java: use package-private (no modifier)
- Kotlin: use
internal (module-visible)
protected is only rarely necessary in either language.
Testing Conventions
BDD Naming
Use BDD-style method names.
Java:
@Test void shouldParseSemanticVersion() { ... }
@Test void shouldReturnEmptyForInvalidInput() { ... }
Kotlin:
@Test fun shouldParseSemanticVersion() { ... }
@Test fun shouldReturnEmptyForInvalidInput() { ... }
BDD Assertions
For verifications, use assertj-core.
Use then(...) instead of assertThat(...):
then(result).isEqualTo(expected);
then(list).hasSize(3).containsExactly("a","b","c");
BDD Mockito
Use given(...).willReturn(...) instead of when(...).thenReturn(...).
Place Mockito given() calls in the "given" block:
given(repository.findById(id)).willReturn(Optional.of(entity));
var result = service.process(id);
then(result).isNotNull();
If the then of AssertJ is imported, too, fall back to verify.
Test Block Formatting
- Always use an empty line to separate given, when, and then blocks (no comments)
- Within these blocks, do not add empty lines between statements
Java:
@Test void shouldCalculateTotal() {
given(taxService.rate()).willReturn(0.1);
var items = List.of(new Item(10), new Item(20));
var total = calculator.calculate(items);
then(total).isEqualTo(33.0);
}
Kotlin:
@Test fun shouldCalculateTotal() {
given(taxService.rate()).willReturn(0.1)
val items = listOf(Item(10), Item(20))
val total = calculator.calculate(items)
then(total).isEqualTo(33.0)
}
Pending Tests
Use @Disabled for pending tests in the TDD test list.
Java:
@Disabled("TODO handle the edge case")
@Test void shouldHandleEdgeCase() { ... }
Kotlin:
@Disabled("TODO handle the edge case")
@Test fun shouldHandleEdgeCase() { ... }
Test Structure
Use @Nested classes to group tests by scenario or method-under-test.
Name nested classes descriptively without a Test suffix, e.g. GivenUserIsLoggedIn.
Use @BeforeEach (and maybe a corresponding @AfterEach) within nested classes
when setup is truly shared across all tests in that group.
If the setup is expensive, use @BeforeAll and a static field within the nested class
(and @AfterAll to clean up).
Keep setup close to tests — only extract to @BeforeEach when multiple tests repeat the same given block.
In Kotlin, @Nested test classes must be declared as inner class (JUnit 5 requires non-static nested classes, and
Kotlin nested classes are static by default).
class OrderServiceTest {
OrderService service;
OrderRepository repository = mock();
@BeforeEach void setUp() {
service = new OrderService(repository);
}
@Nested class GivenOrderExists {
Order order = new Order(42, "pending");
@BeforeEach void setUp() {
given(repository.findById(42)).willReturn(Optional.of(order));
}
@Test void shouldReturnOrder() {
var result = service.getOrder(42);
then(result).isEqualTo(order);
}
@Test void shouldCancelOrder() {
service.cancel(42);
then(order.getStatus()).isEqualTo("cancelled");
}
}
@Nested class GivenOrderDoesNotExist {
@BeforeEach void setUp() {
given(repository.findById(42)).willReturn(Optional.empty());
}
@Test void shouldThrow() {
thenThrownBy(() -> service.getOrder(42))
.isInstanceOf(OrderNotFoundException.class);
}
}
}
Fakes vs Mocks
Neither fakes nor mocks are universally better — choose per dependency.
Fakes are handwritten or simplified implementations (e.g., an in-memory repository backed by a HashMap
or an H2 database).
| Fakes | Mocks |
|---|
| Coupling to implementation | Low — tests verify behavior, not calls | Higher — tests verify specific interactions |
| Readability | High — tests read like production code | Moderate — given/verify adds ceremony |
| Maintenance cost | maintain another implementation | repetition in every test |
| Refactoring resilience | High — internals can change freely | Lower — renaming a method breaks stubs |
| Speed of writing | Slower initially | Faster for one-off tests |
Use fakes when the dependency is stable and used across many tests (repositories, clocks, event buses).
Use mocks when verifying interactions matters, the dependency is volatile, or a fake would be complex to maintain.
If in doubt, start with a Fake.
System Tests
System Tests (STs) go in the test.system package under the test root.
Sub-structure this package as the project grows.
Name ST classes with the suffix ST (e.g., BookResourceST.java / BookResourceST.kt). This
naming is not recognized by Surefire's default includes, so STs are
automatically excluded from normal test runs.
Moving and Deleting Source Files
The Java compiler outputs one .class file per top-level type plus one per nested/anonymous
type or lambda capture, all derived from a single source file. Build tools normally only add new class files;
they never remove stale ones. So if you move or delete a source file, orphaned .class files remain
on disk and can silently shadow the new code or cause other problems.
This also happens with refactoring tools, i.e. a rename does not clean up old class files.
After deleting, moving, or renaming a source file, immediately remove the matching class files from the build output:
find target/classes/com/example \( -name 'Foo.class' -o -name 'Foo$*.class' \) -delete
find target/test-classes/com/example \( -name 'FooTest.class' -o -name 'FooTest$*.class' \) -delete
The glob Foo$*.class covers all nested and anonymous types (Foo$Bar.class, Foo$1.class, …).
Derive the output path from the source path by replacing:
src/main/java → target/classes
src/test/java → target/test-classes
(For multi-module projects, prefix with the module directory as usual.)
IDE Integration
After editing Java or Kotlin files, always check with the IDE for:
- Warnings, not just errors (with IntelliJ you'll have to use
get_file_problems with errorsOnly: false)
- All qualified names replaced by imports (if possible)
- Unused imports removed
Documentation
Java: Use package-info.java files for package-level documentation and architecture.
Kotlin: Use a package.md file (or a dedicated .kt file with only a @file: doc comment and the package
declaration) for the same purpose.
Read these files whenever you read files within a package.
They should help agents to work with the files in a package.