create-cmp

Adversarial audit of one subsystem of a Kotlin/Compose Multiplatform app against its spec AND against platform semantics — the class of defect desktop-tier tests cannot see (alarms, notifications, PendingIntents, reboot, process death, DST). Use this when the user says "audit the notifications", "double check X for bugs", "adversarial review of <subsystem>", "platform audit", "would this survive a reboot", or names a subsystem they want interrogated rather than extended. Reads spec, implementation across ALL source sets and tests; kills its own findings before reporting; lands each survivor in the change flow as a spec amendment + failing-test-first fix or a named human decision — never a direct unreviewed fix to a signed artifact.

Run a create-cmp Compose Multiplatform app's shared UI in a live desktop window with Compose Hot Reload — the daily dev loop. Use this when the user wants to run their CMP app on desktop, preview the app while developing, iterate on Compose UI without an emulator, or asks "run my CMP app on desktop", "hot reload compose", "preview my app while developing", "dev client", "desktop target", "compose hot reload not working", or "see my UI changes live". Covers the hotRunDesktop/run Gradle tasks, how reload triggers, the desktop DI fakes (no Firebase on desktop), and adding desktop fakes for new repositories.

Wire a freshly scaffolded CMP/KMP app to its OWN real Firebase project — the #1 post-scaffold manual step. Use this when the user wants to connect their app to Firebase, or asks "connect my app to firebase", "set up google-services.json", "wire firebase", "create a firebase project for this app", "replace the placeholder firebase config", "get a real GoogleService-Info.plist", "my app still has the REPLACE_ME firebase config", or right after cmp-new as post-scaffold onboarding. Drives the Firebase CLI end-to-end — login, project create/reuse, Android app registration, real google-services.json in place of the placeholder — every cloud-mutating command consent-gated, then proves it with a green assembleDebug. Android-first; iOS branch optional/deferred.

Inspect a running Kotlin/Compose Multiplatform UI as structured design data — hierarchy, geometry and resolved design tokens as JSON, never screenshots — via the cmp-inspector MCP tools. Use this when the user wants to "inspect the Compose UI", "read the design tokens", asks "why is this padding wrong", "check for token drift", "debug this layout without screenshots", "what colour/radius/spacing actually rendered", "did this UI change / regress", "audit this screen for accessibility", "what's on the screen right now", "check the real navigation state", "drive the running app", "tap the app and check", or "prove this UI change did what it should". Tier 0 renders headlessly; tier 1 connects to the RUNNING debug app (real data, real nav state). Asserts on STRUCTURE, not pixels. For a live gallery of every screen, use cmp-preview instead.

Bring up the E2E harness for a Kotlin/Compose Multiplatform app — boot the Android emulator, install the debug build, run the Maestro smoke — or, simpler, run the verify lane, which does all three itself. Use this when the user wants to run E2E/device tests on their CMP/KMP app, "prep my KMP test environment", "smoke test my Compose Multiplatform app", "run maestro on my CMP app", or "verify my CMP app runs on a device". Assumes the toolchain is installed (cmp-doctor) and an app exists (cmp-new).

Generate a regression test suite for a Compose Multiplatform app by OBSERVING it — read the running app's semantics tree as JSON via the cmp-inspector MCP (testTags, text, clickables, bounds, nav state), derive a test plan from what actually rendered, and write the tests into the app's shipped harness (Maestro flows in qa/e2e/*.yaml with testTag selectors, plus spec-cited Compose UI tests). Use this when the user says "write tests for my app", "create a regression suite", "test this screen", "cover this flow with tests", "add UI tests to my CMP app", or "generate tests from the running app". Tests are derived from the rendered structure — never guessed from source, never from screenshots.

Migrate an existing Kotlin/Compose Multiplatform (CMP/KMP) project to the next PROVEN-GREEN dependency version set. Use this when the user wants to upgrade Kotlin, KSP, Compose Multiplatform, Room, AGP, Koin, or Ktor versions in a KMP project, bump their KMP dependencies, or asks "upgrade kotlin/compose/KMP versions", "bump my KMP dependencies", "update my version catalog", "my build broke after updating kotlin", "kotlin and ksp version mismatch", "migrate to the latest CMP", or "is my libs.versions.toml up to date". Diffs the project's gradle/libs.versions.toml against a CI-verified version set, applies surgical in-place edits with backups, guards the kotlin↔ksp lockstep, and can prove the result with the green-build gate. Works on ANY project with a Gradle version catalog — not just create-cmp-scaffolded ones.

Refresh the installed Claude Code plugin from its marketplace and PROVE the result by content — and answer the question a version number cannot: which running sessions have actually loaded which bytes. Use this when a skill behaves like an older version, after publishing a release, when the SessionStart line says the plugin is STALE, or when someone asks "is my plugin up to date", "reload the plugin", "why is my skill out of date", "refresh the marketplace", "did the reload work". The work is a program — `node scripts/plugin-refresh.mjs` — because every hand-run step here has a silent failure that reports success; this skill exists to say when to run it, how to read what it says, and what it deliberately will not do.