Rift

Both programs are tested to 100% coverage, and no test ever reaches the real machine: processes, HTTP calls and waits are all faked.

Rift's tests are written with Pest (opens in a new tab) and cover 100% of app/.

One test file per class, mirroring app/. Commands, and classes that use a Laravel facade, are tested in tests/Feature, the rest in tests/Unit.
Group a class's tests in one describe('method()') per public method, and a command's tests in one describe('when …') per scenario.
Each test checks one behaviour. When a scenario has several effects, give each its own test and share the setup in a beforeEach.
Name the outcome and the condition: it('returns null for a key with an empty value'), it('exits with failure when Docker is not running').
Use datasets for the same behaviour across several inputs.
Nothing reaches the real machine. Fake processes with Process::fake(), the router's admin API with Http::fake() and waits with Sleep::fake().
Do not use @codeCoverageIgnore. A line no test can reach is either dead code or needs a seam.

Command helpers

Command tests build on the helpers in tests/Helpers/Commands.php:

Helper	What it does
`project()`	Create a checkout in a temporary directory and move into it.
`host()`	Fake every process as a healthy machine answers. Overrides are matched first.
`artisan()`	Run a command, ready for output and exit code expectations.
`answerPrompts()`	Let `expectsQuestion()` and `expectsConfirmation()` reply to prompts.

Dashboard helpers

Dashboard tests build on the helpers in tests/Helpers/Dashboard.php:

Helper	What it does
`dashboardSnapshot()`	Build a snapshot of two projects, with a checkout in every state.
`contractSnapshot()`	Build the snapshot `dash/testdata/snapshot.json` is made from.
`collectorHost()`	Fake every process a snapshot reads, for a main checkout and a worktree.
`fakeTerminal()`	Stand in for a terminal that is, or is not, one a person is at.

Coverage reports and TIA

make test:coverage writes an HTML report to .coverage/html/index.html, showing which tests cover each line.

Locally, Pest's TIA engine (opens in a new tab) re-runs only the tests your changes affect, and replays the rest from a cache in .pest/tia.

To	Run
Run every test	`make test:php ARGS="--no-tia"`
Rebuild the TIA cache	`make test:php ARGS="--tia --fresh"`

make test:coverage and CI always run the whole suite, because replayed tests leave the HTML report empty.

rift-dash

rift-dash is held to the same standard, every internal package at 100% coverage, checked by dash/scripts/coverage.sh because go test has no minimum of its own. cmd/rift-dash is left out, as its main only hands over to internal/app.

One _test.go beside each file it tests, with tests named after the behaviour: TestDownAsksThenStops.
The ui package draws screens as plain text. internal/ui/testdata/*.golden holds them; after a deliberate change, rewrite them with make test:go PKG=./internal/ui ARGS=-update and review the diff.
Nothing runs rift or Docker. internal/rift/testdata/fake-rift stands in for rift, printing the contract fixture, and ui.Backend lets tests record what the dashboard asks for.
One test in internal/app runs the whole program on that fake rift, to prove the parts work together.
CI also runs the tests with Go's race detector, which needs cgo and so is left out of the Alpine development image.

make test:coverage writes its Go HTML report to dash/coverage.html.