android-testing-doctrine
Skill1 starsChanged 19 days agoMITView on GitHub
ARC's binding Android testing doctrine — what makes a test real, and the gate every test must pass. Read BEFORE creating any "*Test.kt" file or discussing Android "unit test", "JUnit 5", "JUnit5", "@Test", "runTest", "kotlinx-coroutines-test", "Turbine", "fake", "mock", "MockK", "Robolectric", "ComposeTestRule", "createComposeRule", "screenshot test", "Compose Preview Screenshot Testing", "validateDebugScreenshotTest", "reference images", "coverage", "JaCoCo", "TDD", "golden test vectors", "audit the tests", "tautological test". A test proves observable behavior of production code against an independent oracle. Kotlin/Gradle counterpart of swift-testing-doctrine.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- android-doctrineSkill
- android-verifySkill
- arc-cli-androidSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 56
Never hit a real backend (Firebase included) from a test: Firebase sits behind data-source interfaces.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Tests That Actually Test — Android (ARC Labs)
Precedence. This skill decides whether a test is worth writing and how ARC structures it.
android-skills:testing-setup covers infrastructure mechanics (dependencies, runners, Gradle
setup, screenshot-test plugin vs. AGP test suites). Where it defaults differently — it suggests
JUnit 4 and MockK — ARC's choices below win; otherwise follow it for setup detail.
The single principle
A test verifies observable behavior of production code against an oracle independent of that code (fixture content, external requirement, separate calculation). If the assertion derives from the same line you wrote in production, it is not a test.
Mnemonic: if the test would pass whenever the code compiles, delete it.
Kotlin's compiler already guarantees types, when exhaustiveness over sealed types, nullability
and data-class structure. Blacklist (never write): entries.size == N · data class field
round-trips (Foo(name = "Ana").name == "Ana") · copy()/equals() of data classes ·
mapper round-trips that repeat the mapper · DTO deserialize-and-check-every-field with no
behavior · a UiState constructed in the test and asserted back · Hilt module wiring.
Gate — 5 questions before writing ANY test (all must be yes)
- Does it execute production code (not just construct or read a value)?
- Is the oracle independent of the code under test?
- Could it fail from a behavior change that still compiles?
- Does it assert observable behavior (result/effect/emitted state), not structure?
- Does the seam exist in production by design (interface + constructor injection)?
The ARC stack
| Need | Use |
| Local unit tests (domain, data, ViewModels) | JUnit 5 (mannodermaus Android JUnit 5 plugin) |
| Coroutines | kotlinx-coroutines-test — runTest, StandardTestDispatcher injected through the production dispatcher seam |
Flows / StateFlow<UiState> | Turbine — assert the emitted sequence, not internal fields |
| Test doubles | Fakes over mocks. Hand-written fakes of repository/data-source interfaces in core:testing. MockK only when no seam can be designed — ask first (not on the allowlist) |
| Persistence | Room in-memory database through the real DAO (Room integration on-device per testing-setup when SQLite behavior matters) |
| UI behavior | Robolectric + Compose test (createComposeRule) in test/; semantic matchers first, testTag only when matchers get unwieldy. These run on JUnit 4 rules — keep them on the JUnit 4 runner as testing-setup describes |
| Visual oracle | Compose Preview Screenshot Testing — the same centralized-sample-data @Previews, validated with validate<Variant>ScreenshotTest; reference images updated only on an intended visual change, and the diff reviewed |
| Coverage | JaCoCo (AGP built-in) |
Never hit a real backend (Firebase included) from a test: Firebase sits behind data-source interfaces; tests use fakes. Use an emulator suite only for an explicitly integration-tagged test.
Shared golden vectors with iOS
Pure-logic behavior shared with FavRes-iOS (duplicate detection, statistics, normalization) is pinned by golden JSON test vectors from the shared contract: the same fixture files run in Swift Testing (iOS) and JUnit 5 (Android). The vector is the oracle — never regenerate a vector from the Kotlin implementation's output, and never edit one to make a test pass; a mismatch is a parity bug to raise with the iOS side.
Rules of engagement
- Tests derive from acceptance criteria and are written first (red). A test that passes before the feature exists is wrong — fix it so it fails for the right reason.
- Never modify a test to make it pass — fix the implementation. A change that breaks a previously green test is invalid.
- Nothing is "green" until
./gradlew buildis clean with zero warnings (Kotlin, Lint, detekt, ktlint) — seeandroid-verify. - Auditing an existing suite: apply the Gate to each test; failures get deleted (ask before deleting committed tests); fixtures are kept.
ARC house rules on top
Given-When-Then comments in every test; a makeSut() factory instead of inline construction;
descriptive backtick names (`emits Error when repository throws`); JUnit 5 @Tags
unit / integration / ui / critical; @ParameterizedTest for repeated logic. Coverage
floor: apps 80% (packages 100%). Coverage is a floor, not a goal — a tautological test
that lifts the number still fails the Gate and still gets deleted.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

