Android Project Standard

Android Project Standard

This skill is the guiding standard for any Kotlin/Android work. Apply it by default; don't wait to be asked.

Its spine: trust is placed not in the model, but in the machine-checkable code that constrains it. In Kotlin, a large part of that scaffold is the compiler — null-safety, sealed when exhaustiveness, and (with explicitApi()) an explicit public API surface are enforced at compile time. But Kotlin lacks Rust's Send/Sync or Swift 6's compile-time data-race safety, so structured concurrency discipline + detekt's coroutine rules backstop it. The job here is to (a) keep the escape hatches shut and turn on allWarningsAsErrors + explicitApi() + a detekt error-gate, (b) push every external dependency behind an interface so the model is hot-swappable, (c) organize deep with module-per-domain and name-navigable structure, and (d) mechanize the implicit knowledge a human would otherwise hold — via a zero-warning gate, a per-module contract, and the two lint gates. The agent's output ceiling equals the tightness of that loop.

Baseline: Kotlin 2.x (kotlin("jvm")/com.android.application), allWarningsAsErrors = true at every module, explicitApi() on library modules, detekt (io.gitlab.arturbosch.detekt, escape-hatch rules as error) + ktlint (org.jlleitschuh.gradle.ktlint, format), kotlinx.serialization + @JvmInline value class newtypes, a version catalog (gradle/libs.versions.toml) pinning everything, Hilt + Jetpack Compose + Material3 in :app, KSP for annotation processing.

This standard is the project-level engineering charter (the gate + architecture conventions). It is complementary to polyglot-core-standard: that meta-standard governs the cross-language seam (one canonical Rust core + derived UniFFI/JNI bindings that are vendored, linguist-generated, gate-excluded, and never edited; a typed/fallible FFI boundary; a composed gate). This skill governs the Kotlin host's internals — lint, format, strict compilation, module structure, the provider seam, and tests. In a polyglot repo the two compose: polyglot defines "how the seam connects", this defines "how Kotlin is written inside", and this skill's ./ci.sh plugs into polyglot's make check-android slot (today the weakest cell of that composed gate — just gradlew compileDebugKotlin).

When starting a new project

python scripts/scaffold.py com.example.notes --target <dir> --domains retrieval generation

This mirrors assets/templates/ into a Gradle multi-module project (settings.gradle.kts + :kernel/:domain/:adapters library modules + a :app Android application + a kotlin("jvm") module per --domains + a CLAUDE.md per module + ci.sh + ADR), substitutes __APP_ID__ in file contents and rewrites the __PKG_PATH__ source-package directory, and registers each domain module in settings.gradle.kts + :app dependencies. Module names (:kernel, :domain, …) are project-independent; only the app id, namespace, and CLAUDE.md headers carry the literal name. When adapting an existing repo, copy from assets/templates/ by hand.

Then: gradle wrapper --gradle-version 8.11.1 (once), and ./ci.sh (or make check) to verify. detekt + ktlint are wired via Gradle plugins; no extra install.

When working on existing code

Apply the rules below, keep new code strict, and verify structural invariants:

python scripts/check_conformance.py <project_root>

It parses settings.gradle.kts / build.gradle.kts (text/regex — Gradle has no clean manifest-dump) and checks the mechanically enforceable invariants (multi-module present, :domain zero-SDK + zero-framework + no :adapters dependency, allWarningsAsErrors wired per module or in root, detekt + ktlint config present, detekt.yml marks !!/UnsafeCast as error with a fail threshold, generated bindings gate-excluded, ci.sh chaining ktlint → detekt → compile(-Werror) → test, a CLAUDE.md in every module). Everything else is enforced by the gate and by applying the standard.

The non-negotiables

Full rationale in references/standard.md.

1. The compiler is the static guarantee; turn on the strict knobs and keep the escape hatches shut. Every module sets allWarningsAsErrors = true; library modules add explicitApi() (public API must be explicit). The escape hatches are banned as detekt errors: no !! (UnsafeCallOnNullableType), no unchecked as (UnsafeCast → use as?), no lateinit abuse (LateinitUsage), no platform-type leaks, no GlobalScope (GlobalCoroutineUsage). A site that genuinely needs one takes the controlled exit: @Suppress("UnsafeCallOnNullableType") // reason (explicit, auditable, greppable — the Kotlin analog of a justified // swiftlint:disable:next … — reason).

2. No silent failures: a sealed error hierarchy / Result<T>, not swallowed exceptions. Recoverable failures are a sealed ProviderError/DomainError (exhaustive when) or Result<T>; vendor/network/timeout errors normalize at the adapter boundary into a domain error; program bugs trap (error(...)/require(...)) and propagate — never swallowed (detekt SwallowedException = error). The Kotlin mirror of Rust's Result<T,E> + thiserror and Swift's typed throws.

3. Boundaries: parse, don't validate — kotlinx.serialization + value-class newtypes. Everything crossing a boundary (config, LLM output, tool results, files) decodes into a strongly-typed value via kotlinx.serialization; constraints are encoded in @JvmInline value class newtypes with a validating factory (of(...) that throws), so invalid states are unrepresentable. A custom serializer routes decoding through the factory. The Kotlin mirror of pydantic / serde + newtype.

4. Model-agnostic: every external AI dependency behind an interface in :domain; SDKs only in :adapters. The :domain module has zero SDK and zero framework dependencies (the checker enforces this); it is a pure kotlin("jvm") library. Real SDKs live in a separate gated module (:adapters-openai, conditionally included by a Gradle property — the analog of Cargo features / SPM traits) and normalize to ProviderError. The composition root (:app's Hilt @Module) is the one place that selects an impl by config and injects it via @Provides. A deterministic MockProvider is the default (not a test stub) so the app, tests, and CI run offline with no SDK or key.

5. Completion = one zero-warning gate, the agent's only correctness judge. ./ci.sh runs ktlintCheck → detekt → compileDebugKotlin (with allWarningsAsErrors) → test (offline, Mock default, smoke + conformance), under set -euo pipefail. Run it after every change; fix until green; never "looks fine, commit". Pin it with a pre-push hook; CI mirrors it; in a polyglot repo it plugs into make check-android.

6. Two lint gates, both in the gate. ktlint owns formatting (the swift-format/cargo fmt analog, via .editorconfig); detekt owns lint and the escape-hatch bans (the SwiftLint/clippy analog, via config/detekt/detekt.yml with maxIssues: 0). They are complementary, not redundant — keep both, both in the gate.

Structure: Gradle multi-module, module-per-domain

The depth comes from the modules, not from one module with deep folders. A Gradle module ≈ a Cargo crate / SPM target: module-level dependency isolation. "Fix the reranker" should resolve to :retrieval/... with no search.

• A module per bounded context, plus the fixed infra modules and a thin app:
  • :kernel — cross-cutting infra: Config.kt (@Serializable AppConfig; defaults < configs/settings.json < env APP_* with __ nesting), Log.kt (platform-agnostic Log + logProvenance; payloads never logged), Prompts.kt (classpath resources + strict rendering). kotlin("jvm"), zero external deps.
  • :domain — ports (interfaces) + models (value-class newtypes) + sealed boundary errors; zero SDK, zero framework.
  • :adapters — interface impls; MockProvider is the default; ProviderFactory is the assembly seam; real SDKs in a gated sibling module.
  • domain feature modules (:retrieval/:generation/…) depend on :domain + :kernel, never on :adapters or SDKs.
  • :app — com.android.application; Compose UI + the Hilt composition root that wires concrete adapters by config.
• Dependency direction (one-way): feature / :adapters → :domain + :kernel; :app → everything. :domain is zero-SDK and never depends on :adapters.
• A CLAUDE.md in every module: the root one is a routing table (hard constraints + where to look + the polyglot boundary); each module's states its responsibility, dependency direction, and local contract (:domain's "zero SDK", :adapters' "errors normalize", :kernel's "payloads never logged"). The checker requires one per module.
• Go deep: split a module into sub-packages by sub-capability; names map to paths.

Navigability

Naming-as-path is to navigation what types are to interface contracts — and Gradle lifts it to the module level. Group by capability, not by models//utils/. Nest until leaf files have a single clear responsibility. A name should resolve to a path with no search.

Principles for AI-touching code (advisory)

Beyond the type system — for any code where a model produces output. Upper-level discipline; not all mechanically checkable.

• Constrain, don't ask. Push non-negotiable properties (no fabrication, must-cite, no privilege escalation) into deterministic control flow so the model physically cannot violate them — don't rely on the prompt. Synthesize answers from typed values in code; discard model prose on the critical path.
• Narrow the emission surface. Make the model pick from a typed sealed/enum or call tools returning a tri-state result; take final values from tool results, not free-form model text. A Kotlin sealed class + exhaustive when is the natural controlled surface.
• Guardrails are deterministic, independent, never pluggable. Intent parsing can be swapped; safety decisions are deterministic code re-evaluated from raw input, not trusting a pluggable component's output. (In a polyglot repo these live in the core, not the host.)

Driving AI on big work (advisory)

Treat AI as supervisable labor, not unsupervised autopilot.

• Decision-first: a numbered, immutable ADR (docs/adr/) before coding — context + chosen option + rejected alternatives and why. Rejected-reasons stop it re-walking excluded paths.
• TDD red-light first; tests are the immutable spec. Write the failing test (JUnit / kotlin.test), then implement to green; never weaken a test to pass.
• Numbered steps, each independently green; one commit per step through the full gate. No giant diffs. (In a polyglot repo, a core-contract change + regenerated bindings + the host adaptation is one coherent commit.)
• Adversarial independent review. After writing, run a separate, hostile, multi-perspective review prioritizing what tests can't cover (diagrams, docs, tradeoffs).

Scale to project size

The module split, per-module CLAUDE.md, ADRs, and provider seam are real overhead — overkill for a 200-line app. Present them as triggered, scalable patterns ("the moment you call an LLM/embedding/vector store, put it behind an interface in :domain"; "split a module out the moment it takes a second responsibility"), not blanket mandates. A single-module app + the two lint gates + the zero-warning gate is a perfectly good small-scale instantiation.

Scale along two axes, not one — size and domain. The standard has a universal spine that holds for any Kotlin/Android project whether or not it touches AI: Kotlin compiler strictness + allWarningsAsErrors + explicitApi, the zero-warning gate, sealed errors / Result with the escape hatches shut, kotlinx.serialization + value-class newtypes at boundaries, Gradle module-per-domain + deep naming-as-path, a version catalog, ktlint + detekt. The rest is an AI-triggered layer that only switches on once the project actually calls an LLM / embedding / vector store: the :domain interface seam + zero-SDK module, MockProvider-as-default, bundled prompts + strict rendering, logProvenance, and the constrain-don't-ask discipline. A pure app that never calls a model should take the spine in full and skip the AI layer outright — bolting MockProvider onto such a project is cargo-culting, not conformance. (In a polyglot repo the AI seam lives in the one sub-tree that calls the model — usually the core — and the Kotlin host stays spine-only, receiving typed values across the FFI boundary.)

Resources

• references/standard.md — the full standard with rationale and complete code for every module (read for the why, edge cases, or exact module contents), plus the mapping to the Python/Rust/Swift sibling standards and the polyglot seam boundary.
• assets/templates/ — exact Gradle boilerplate; __APP_ID__ (content) + __PKG_PATH__ (source-package dir) are the only placeholders. Includes :kernel/:domain/:adapters/:app, detekt.yml, .editorconfig, ci.sh, Makefile, a per-module CLAUDE.md, gradle/libs.versions.toml, configs/settings.json, and smoke/conformance tests.
• scripts/scaffold.py — generate a conforming multi-module project.
• scripts/check_conformance.py — verify structural invariants (incl. :domain-zero-SDK and per-module CLAUDE.md).