AGENTS

Cluster-E2E Agent

End-to-end HMI automation from Figma design to QNX runtime — static generation, signal adapter, PC simulation loop.

Core Pain Points

1. Scattered requirements, designs, and assets: PRD/SRD, Figma/UI annotations, UE assets, and sliced image resources live in different sources.
2. Heavy page setup + asset binding: Page layout, component hierarchy, motion, fonts/colors, and asset binding require repeated manual configuration.
3. Long integration chain, slow issue localization: Multiple rounds of joint debugging between HMI static generation, QNX logic layer, and PC simulation; when differences appear in build, rendering, or screenshots, localization and correction depend on humans.

Solution & Technical Implementation

1. Standardized inputs + structured intermediate artifacts: Accepts PRD/SRD, Figma/UI annotations, UE assets, and sliced images; generates JSON/CSV contracts such as page_layout, asset_index, node_hierarchy, and runtime_state.
2. HMI static generation: Converts Figma and other inputs into structured JSON; completes flat-node → tree hierarchy construction, Prefab/Component generation, motion application, and visual verification.
3. QNX logic + signal adaptation: A three-layer framework — signal adaptation layer, logic-processing layer, signal-sending layer — converts CAN/SOME-IP into internal signals; after state-machine and business-rule processing, drives the UI via Unity NativeDataControl.
4. PC simulation + closed-loop correction: Web console → WebSocket → RuntimeStateBridge → runtime_state.json → DashboardController drives Unity updates; after build, render, and screenshot, compares against the design draft to complete difference attribution and automatic correction.

Highlights

• Multi-skill collaboration: Includes 8 capabilities: hmi-unity-generator, hmi-expected-compositor, hmi_ui_end_to_end, hmi-sim-architect, image-comparer, and others.
• Structured contracts decouple modules: Intermediate artifacts can be repeatedly consumed by skills, versioned, and traced.
• Unity as a shared convergence point: HMI static output, QNX real-machine logic, and PC simulation collaborate around the same project.
• QNX/WIN dual platforms + closed-loop error correction: Addresses signal adaptation, simulation execution, and automatic correction of visual differences.

Business Value

• Manual cycle ~3 months, E2E Agent ~1 month; manual effort ~12 person-months, E2E Agent ~3.5 person-months, covering the full HMI feature set.
• Structured inputs, Unity projects, and QNX/PC signal-adaptation layers are reusable, adapting to different types of HMI projects.
• The rollout path is divided into four phases: MVP, expansion, dual-engine, and compliance delivery — evolving from demo to mass production.

---