Can an AI Lobster Build Cross‑Platform Apps? A Hands‑On Kuikly App Builder Walkthrough

Motivation

Client‑side development traditionally relies on heavyweight IDEs (Xcode, Android Studio, DevEco‑Studio). As AI agents become the primary consumers of software tooling, the lack of command‑line interfaces (CLI) and structured pipelines prevents AI from automating project creation, compilation, and debugging in the same way it can for front‑end or back‑end ecosystems.

Kuikly Framework Overview

Kuikly provides native rendering combined with Kotlin‑Multiplatform (KMP) to enable a single codebase to run on Android, iOS, HarmonyOS, and H5. The framework supports dynamic module loading and cross‑platform UI components, but project scaffolding historically required a GUI plugin inside Android Studio, which blocks programmatic access.

Kuikly App Builder Skill

The "Kuikly App Builder" skill equips an AI agent with three core capabilities:

CLI‑based project lifecycle : A single command creates a complete Kuikly project, generating shared modules and host applications for Android, iOS, HarmonyOS, and H5 in one step. Example command (illustrative):

kuikly-app-builder create --name MyApp --platform android,ios,hm,h5

Structured JSON output for all commands : Compilation failures are parsed into JSON objects that include the file path, line number, error message, and suggested fix. Sample JSON schema:

{
  "error": "CompilationFailed",
  "file": "src/commonMain/kotlin/com/example/App.kt",
  "line": 42,
  "message": "Unresolved reference: Foo",
  "suggestion": "Import com.example.Foo"
}

This enables the AI to locate the problem, modify the source, and re‑invoke the build without human intervention.

Knowledge injection via Skills file : Component APIs, common pitfalls, and version‑compatibility matrices are packaged into a Skills repository. The AI is forced to consult this documentation before generating code, avoiding incorrect assumptions from other frameworks (e.g., React or Flutter). The Skills repository is open‑source at https://github.com/Tencent-TDS/KuiklyUI-AI.

Implementation Details

During skill execution the AI performs the following steps:

Invokes the CLI to scaffold the project.

Runs the build command ( ./gradlew assembleDebug for Android, xcodebuild for iOS, etc.).

Parses the build log into the JSON format described above.

Applies the suggested fixes automatically and repeats the build until success.

The Skills file also contains rules such as "prohibit writing code from memory"; the AI must first fetch the official Kuikly API reference (available at https://kuikly.tds.qq.com/AI/) before emitting any source files.

Results

A functional stock‑market application was generated entirely by the AI. The app performs real network requests using a provided API key, builds an APK, and is automatically deployed to Tencent’s internal cloud‑device platform for remote execution and testing.

Two current limitations were observed:

The cloud execution environment sometimes lacks the toolchain required to compile iOS binaries, preventing full multi‑platform verification.

The workflow implements only a minimal closed loop (create‑build‑fix). It does not yet cover end‑to‑end demand definition, acceptance testing, or continuous harness engineering.

Key Takeaways

Exposing a framework through a CLI, providing machine‑readable error output, and documenting its programming model are prerequisite steps for AI‑agent friendliness.

With these foundations, AI agents can already produce prototype‑level cross‑platform applications; extending the skill set to include debugging, performance analysis, and development‑pattern management will further narrow the gap with human experts.

Technical References

Kuikly open‑source repository: https://github.com/Tencent-TDS/KuiklyUI Official documentation:

https://kuikly.tds.qq.com/Introduction/arch.html