How Huawei Cloud CodeArts (Py4OH) Rebuilt OpenHarmony Python Development with Spec‑Driven AI Automation

Project Background – Py4OH, a Python‑for‑OpenHarmony development suite created by the Electronic Science University OpenHarmony club and the ShuHong community, suffered from outdated build environments, difficult debugging, tangled code quality, and lack of engineering standards.

Why Rebuild with Huawei Cloud CodeArts – The team chose Huawei Cloud CodeArts (CodeArts) to start from a clean slate, using AI agents to extract functional semantics from the 9 million‑line legacy code, discard debt, and regenerate equivalent, modern code.

Spec‑Driven Development (SDD) – Requirements are captured in spec.md files that define "what to do" and "how to do it". AI agents translate these specs into concrete implementations (e.g., CreateCustomBlockModalOptions) and then human reviewers verify the output, forming a three‑layer collaboration loop.

AI‑Powered Code Generation & Review – The AI automatically detects issues such as calling dispose() after render(), missing default values, or silent parameter deletions, fixing them before commit. It also performs static analysis (TypeScript strict mode, ESLint) and runtime safety checks (dangerous character regex, path traversal protection, input length limits).

Automated translation of specifications to code reduces manual effort.

AI‑assisted review catches hidden bugs early, improving code quality.

Modular code generators ( blocks‑codegen‑refactor) split monolithic generators into category‑based modules, enabling on‑demand loading.

Cross‑Platform Bridge (Py4OHBridge) – A singleton bridge detects the runtime environment and selects the appropriate client: SSE for web, IPC for Electron. It provides unified commands ( Py4OH‑REPL, notifyReady) and automatic reconnection.

Security & Internationalization – All commands pass through Py4OHCommandBuilder, which enforces three layers of safety (dangerous character regex, path traversal, null byte checks). UI strings are managed via i18n files ( zh‑CN.ts, en‑US.ts) with enum keys (e.g., BlockTextKey.CUSTOM_EDIT), ensuring consistent multilingual support.

Full‑Chain Packaging – The build pipeline uses Docker/Caddy for secure headers, SSE for server‑client sync, and Electron IPC for native packaging, with AI agents providing cross‑layer dependency analysis during packaging.

Outcome – Py4OH‑2.0 achieves an 80 % reduction in code size (≈2 kB vs. 10 kB), supports rapid low‑code development, and enables a weekly 2‑3‑feature delivery cadence while maintaining high code quality, security, and architectural consistency.