KMP Generate All Tests

This documentation was generated with the assistance of AI. Please report any inaccuracies.

Bring one Kotlin Multiplatform module’s unit test coverage up to a minimum bar — 80% line coverage, measured on the JVM/Android host runs — in one pass: find every class with no test at all or with measured coverage below the bar, then write or extend tests for each until it clears it. Kotlin Multiplatform/Gradle only; it never runs against, or falls back to, another stack.

Purpose

iru-kmp-generate-all-tests answers "bring this whole module’s test coverage up to a bar" rather than "write tests for this one class/task", which iru-kmp-code-one-task and the iru-code pipeline already handle per task. It picks the module (default shared, or composeApp in the wizard layout, or androidApp when scoped to it), uses the Explore agent to inventory every production class with its source set and the module’s existing test conventions, and delegates to iru-kmp-coverage (through iru-gate-runner) for the exact, measured percentage per class.

Each test goes in the source set matching the code under test — commonTest for commonMain (running on every target), the Android host-test set for androidMain, iosTest for iosMain, androidApp/src/test for the app — defaulting to kotlin.test + runTest + Turbine (when present) + hand-written fakes (MockK never in commonTest; Mokkery only if already applied). Kotlin/Native has no coverage tool, so iosMain/appleMain/nativeMain classes are "not measurable", get behavior tests only when invoked with ios: yes, and never count against the bar. It adds only test code; production code is touched only when a class is genuinely untestable as written, and that is reported explicitly. It stops immediately if the repository is not a Kotlin Multiplatform Gradle project, or has no JVM test target to measure.

When the module renders UI and its tests are already wired for it (a commonTest uitest/UiScreenshots.kt helper with the Robolectric UiTest base for ui: compose, or androidApp/src/test/…​/uitest/UiScreenshots.kt and iosApp/iosAppTests/UIScreenshots.swift for ui: native), the new Step 5a also generates UI tests per the catalog’s UI contract (see Building user interfaces) for every existing screen and component that has none. One agent per missing screen/component writes a test that asserts each event once and captures every visual state (minimal, typical, many-items/long-content and right-to-left fixtures) across the .ui-quality.yml directions × themes into .ui-screenshots/ with the --android/--ios platform segment, plus the accessibility check. The tests then run through iru-kmp-test’s `ui selector (iOS parts unverified without macOS or a simulator). Without the wiring the uncovered UI is listed as "needs the UI test wiring" — the skill never adds Robolectric, a dependency or a test target.

Purpose group

Kotlin Multiplatform Development Tooling

Plugin

Shipped by the iru-kmp plugin of the ai-catalog marketplace (install with /plugin install iru-kmp@ai-catalog; see Plugins).

Invocation

/iru-kmp-generate-all-tests [scope]

Inputs

Argument Required Description Default

scope

No

One module (:shared, androidApp) or a comma-separated list of classes within it; each class’s module and source set are resolved from its file path. Also accepted as a scope: line in args.

shared (or composeApp when it is the only KMP module); asked via AskUserQuestion when several KMP modules exist and none is named.

ios

No

yes or no, passed as an ios: line in args. With yes, iosMain classes lacking an iosTest test get behavior-only tests, and the final suite run includes the iOS stage.

no

Outputs

  • New or extended test files in the source set matching each class: commonTest, the Android host-test set (androidHostTest or androidUnitTest), iosTest (only with ios: yes), or androidApp/src/test.

  • Occasionally, a minimal testability change to a class under test (e.g. extracting a common interface and injecting it) — only when a class is genuinely untestable as written, and always called out explicitly.

  • A report: module scoped, candidate count and classes excluded as having no testable logic, conventions followed, how many were already at or above 80% and how many iOS-only classes were not measurable, before/after line coverage per class touched, any class still below 80% after two extension attempts (with uncovered lines and likely reason), and the full-suite result per target including any unverified iOS stage.

  • For UI modules wired for it (Step 5a): UI test classes for every screen and component that lacked one, the .ui-screenshots/ captures they write, and a report of captures per kind and platform, accessibility violations (count + rule ids) with the .ui-quality.yml level, unverified parts, and UI left uncovered because the wiring is missing.

  • No license headers, KDoc, code-quality fixes, on-device Android tests, or iosApp XCTest tests unless explicitly requested (the ImageRenderer UI tests of ui: native modules wired for them are the exception, via Step 5a).

Execution flow

flowchart TD A[Start /iru-kmp-generate-all-tests] --> B["Step 0: parse scope and ios args"] B --> C{"KMP module and\nGradle wrapper found?"} C -- no --> C1[Stop: Kotlin Multiplatform\nGradle only] C -- yes --> D["Step 1: pick module (ask if several);\nlist Gradle tasks; note host capability"] D --> D1{"JVM test target\nexists?"} D1 -- no --> D2[Stop: coverage needs a host-test\nsource set] D1 -- yes --> E["Step 2: Explore agent inventories every class,\nits source set, and the test convention"] E --> F["Step 3: gate-runner runs kmp-coverage once\nover every candidate (batched if large)"] F --> G["Step 4: classify - missing tests / below 80% /\nsufficient / not measurable (iOS)"] G --> H["Step 5: one agent per class (parallel batches)\nwrites or extends tests in the matching source set;\ncompile-check touched test source sets"] H --> H2["Step 5a: UI module with test wiring? agent per\nscreen/component without UI tests; run via kmp-test ui"] H2 --> I["Step 6: re-measure via kmp-coverage"] I --> J{"Still below 80%?"} J -- yes, attempts left --> H J -- yes, 2 attempts used --> K[Report as residual gap] J -- no --> L["Step 7: gate-runner runs kmp-test full suite\n(android, plus ios when in scope)"] K --> L L --> M{"Failures caused\nby this run?"} M -- yes --> N[Fix regression; re-run] N --> L M -- no --> O["Step 8: final report"]

Dependencies

Invokes

  • KMP Coverage — Step 3, and again in Step 6 to re-measure after generation: delegated through the iru-gate-runner agent so the Kover report never lands directly in this conversation and the JVM suite runs once per batch rather than once per class.

  • KMP Test — Step 7, once, to confirm the full suite still passes (Android, and iOS when ios: yes or a commonTest file changed on a macOS host with Xcode); delegated through iru-gate-runner, which lets the skill take .integration-test.lock itself for any iOS stage. Step 5a also invokes it with the ui selector to run the newly generated UI tests.

Invoked by

None from inside this catalog — it is a top-level entry point invoked directly by a person who wants coverage brought up across one KMP module in one pass, independent of the /iru-code//iru-issue pipeline (which handles per-task test-writing via iru-kmp-code-one-task). iru-android-generate-all-tests only points to it in prose when it detects a Kotlin Multiplatform project, and iru-kmp-coverage mentions it as a possible consumer of its report.

Related agents

  • Explore — Step 2, to build the class/test-file inventory across the module without reading every source file directly.

  • iru-gate-runner — Steps 3, 5a, 6, and 7, to run iru-kmp-coverage/iru-kmp-test and report back only a compact summary instead of a raw coverage/test log.

  • general-purpose (default, no subagent_type named) — Step 5, one per class needing new or extended tests, dispatched concurrently since each only touches its own test file; Step 5a also uses it for one agent per screen/component needing UI tests.

Source

SKILL.md on GitHub — the file this page was generated from.