Writing tests with Testo

The attribute set, assertion facade, exception expectations, and lifecycle hooks are Testo's own. Write them the Testo way described below — don't transliterate idioms from other test frameworks.

Before you write code

Fetch the canonical API surface:

• https://php-testo.github.io/llms.txt — concise index. Always start here.
• https://php-testo.github.io/llms-full.txt — escalate when llms.txt doesn't answer the question.

If the project ships an AGENTS.md, honour it.

Canonical shape of a test class

<?php
declare(strict_types=1);

namespace Tests\Unit;

use Testo\Assert;
use Testo\Codecov\Covers;
use Testo\Test;
use App\UserService;

#[Test]
#[Covers(UserService::class)]
final class UserServiceTest
{
    public function createsUserWithGivenName(): void
    {
        $service = new UserService(new InMemoryRepository());

        $user = $service->create('Alice', 'alice@example.com');

        Assert::same($user->name, 'Alice');
    }
}

Hard rules:

• Class-level #[Test] when every public method is a test (preferred). Method-level #[Test] when only some are.
• final class by default.
• No base class — Testo does not require one.
• Public methods returning void or never under a #[Test] class are auto-discovered as tests.
• One #[Covers(...)] at class level when all tests cover the same class; at method level when they differ.
• Arrange / Act / Assert separated by a single blank line. Do not write // Arrange, // Act, // Assert comments.
• File path mirrors the source: src/Foo/Bar.php → tests/Unit/Foo/BarTest.php (or wherever the suite finder is rooted).

Assert facade (immediate checks)

Use the Testo\Assert facade for in-test checks. Order is actual, expected for same/equals.

Assert::same($user->id, 42);
Assert::notSame($a, $b);
Assert::equals($result, '1');         // loose ==
Assert::true($flag);
Assert::false($flag);
Assert::null($value);
Assert::blank($value);                // null, '', [], or 0-count
Assert::notBlank($value);             // inverse of blank(); false/0/'0' count as non-blank
Assert::contains($collection, $needle);
Assert::count($collection, 3);
Assert::instanceOf($object, MyClass::class);
Assert::fail('explicit failure');

Typed chains (use when you want a fluent series of checks on one value):

Assert::string($s)->contains('foo')->notContains('bar');
Assert::int($n)->greaterThan(0)->lessThanOrEqual(100);
Assert::numeric($n)->between(1, 100);  // int, float, or numeric string
Assert::array($a)->hasKeys('id', 'name')->isList()->hasCount(3)->contains('x')->notContains('y');
Assert::array($a)->sameElementsAs([3, 2, 1]);  // order-insensitive, keys ignored
Assert::object($o)->instanceOf(Foo::class)->hasProperty('id');
Assert::json($s)->isObject()->hasKeys(['data', 'meta'])->assertPath('$.data.id', 42);

Expecting exceptions

Use Testo\Expect declared before the Act phase. The test method's return type is never.

use Testo\Expect;

#[Test]
public function rejectsNegativeAmount(): never
{
    Expect::exception(InvalidArgumentException::class)
        ->withMessage('amount must be positive')
        ->withCode(1001);

    new Account(-100);
}

Other Expect modifiers: withMessageContaining(...), withPrevious(class, closure), memory-leak expectations. Do not use try/catch-based assertions for expected exceptions — Expect::exception is the correct API.

Marking a test as skipped or cancelled

Throw a status-bearing exception from the test body to short-circuit the run with a non-error verdict:

use Testo\Core\Exception\SkipTest;
use Testo\Core\Exception\CancelTest;

#[Test]
public function requiresPdoMysql(): void
{
    if (!extension_loaded('pdo_mysql')) {
        throw new SkipTest('pdo_mysql required');
    }

    // ... real test ...
}

• SkipTest → Status::Skipped. Use when the test isn't applicable in this environment (missing extension, disabled feature flag, unavailable optional dependency, etc.).
• CancelTest → Status::Cancelled. Use for cooperative cancellation (deadline expired, Fiber unwind). Not a generic "I don't want to run" — that's SkipTest.

Constraints:

• Must escape the test method itself. The runner's inner try/catch maps the throw to a status; raising from an interceptor or #[BeforeTest]/#[AfterTest] hook bubbles out of the pipeline and is treated as Status::Aborted instead. To skip from a hook, leave the precondition check inside the test body.
• These are not assertions — don't try/catch them inside the test, just throw.
• Subclasses work: class MissingExtensionSkip extends SkipTest {} is still recognized.
• Return type stays void, or never if the throw is unconditional.

Tests that intentionally perform no assertions

A test that finishes successfully without recording a single assertion is reported as Status::Risky — the framework assumes you forgot to assert. When a test legitimately verifies behaviour without the Assert facade (e.g. it only checks that a call does not throw), declare that intent with #[ExpectNoAssertions] to keep it Status::Passed:

use Testo\Assert\ExpectNoAssertions;

#[Test]
#[ExpectNoAssertions]
public function bootsWithoutError(): void
{
    (new Kernel())->boot();   // success is simply "no exception thrown"
}

Place it on a single test — a method or a function. It is not allowed on a class: "no test here asserts anything" is rarely a real contract, and a stray class-level marker would flip every genuinely-asserting test to Risky.

The attribute is a two-way contract, not just a switch: a marked test that does record an assertion is reported as Status::Risky (the declaration is stale or wrong). This includes Expect::exception(...) / #[ExpectException] — expecting an exception is itself an assertion, so pairing it with #[ExpectNoAssertions] is contradictory and comes out Risky. Use the attribute only on tests that truly assert nothing.

#[ExpectNoAssertions]	test records an assertion	status
no	no	Risky (forgotten assertion)
no	yes	Passed
yes	no	Passed
yes	yes	Risky (stale/misapplied attribute)

Lifecycle hooks

use Testo\Lifecycle\{BeforeClass, AfterClass, BeforeTest, AfterTest};

#[BeforeClass]
public static function bootSchema(): void { /* once before any test */ }
#[BeforeTest]
public function openTx(): void           { /* before each test */ }
#[AfterTest]
public function rollback(): void         { /* after each test */ }
#[AfterClass]
public static function dropSchema(): void { /* once after all tests */ }

Hooks may be either instance methods or static — Testo invokes them accordingly. They run regardless of #[Test] on the method.

In a function-based test case (a file of top-level #[Test] functions rather than a class), the same attributes work on plain functions. The hooks apply to that file's case — #[BeforeClass]/#[AfterClass] run once around the whole file, #[BeforeTest]/#[AfterTest] around each test function. A lifecycle function needs no #[Test] and is never itself a test; share state through a static holder, since functions have no $this.

use Testo\Lifecycle\BeforeTest;
use Testo\Test;

#[BeforeTest]
function openTx(): void { Db::$tx = Db::begin(); }   // before each test function in this file

#[Test]
function insertsRow(): void { /* ... */ }

Grouping tests

Label tests with #[Group] (from the testo/filter plugin) to select or skip them by category. It targets classes, methods, and functions and is variadic (pass several names at once).

use Testo\Filter\Group;

#[Test]
#[Group('driver-mysql')]     // inherited by every test of the class
final class MysqlConnectionTest
{
    #[Group('slow')]         // effective groups: driver-mysql, slow
    public function importsLargeDataset(): void { /* ... */ }
}

A test's group set is the union of all groups reachable from it: its own method (and any overridden parent method), the test class, its parent classes, and traits. Groups are selected at run time with --group — see the testo-run-tests skill.

Running

Run the test you just wrote through the Testo CLI, always with --json:

vendor/bin/testo --json --filter='UserServiceTest'

Filter selection (--suite/--filter/--path/--group/--type), the JSON report shape, and exit semantics are covered by the testo-run-tests skill — escalate there before adding other flags.

Pitfalls

• Do not mock enums or final classes — instantiate real ones.
• Do not invent attributes. If you need behaviour you haven't seen in llms.txt, escalate to llms-full.txt before guessing.
• Do not write setUp/tearDown — use the lifecycle attributes above.
• For parameterized tests, escalate to the testo-data-driven skill.
• For flaky-test handling, escalate to the testo-flaky-tests skill.
• For fiber/coroutine or async I/O tests (\Fiber::suspend(), amphp, Revolt, Future::await()), escalate to the testo-async skill.
• For exception assertions, always use Expect::exception(...) before the throwing call — never wrap in try/catch.