AGENTS

DEV Agent

Design policy generation + Block diagram component impact analysis for automotive ECU changes.

Core Pain Points

1. Design-assessment guidelines depend heavily on senior engineers: Requires simultaneous understanding of change points, historical design cases, design guidelines, methods, evaluation methods, and concerns.
2. Historical case matching relies on keywords, easily misses recalls: Wording of old and new changes may differ; synonym matching is difficult.
3. Number correspondence across four guideline categories is hard to maintain: Design guideline → design method → evaluation method → concern must correspond one-to-one by number; manual arrangement is prone to misalignment.
4. Understanding bias in multilingual automotive terminology: Change descriptions may mix Chinese, Japanese, English, abbreviations, and code.
5. Block-diagram impact analysis relies on system-architecture knowledge: Dashboard dual-chip architecture means impact judgment involves processing units inside functional modules.
6. Manual editing of Drawio XML is error-prone: Each component corresponds to one mxCell; manual edits require locating cell_id and updating styles.

Solution & Technical Implementation

1. Receive change points and unify language handling: The required input is changing_point; non-Chinese input is converted to Chinese for knowledge retrieval, while output preserves the user's original language.
2. Retrieve relevant designs from historical knowledge: Retrieve the top-3 most relevant design references from knowledge.json via keyword extraction → Neo4j CONTAINS/BM25 → Top-3, with fallback to n-gram matching.
3. Generate structured design-assessment guidelines: Uses automotive-electronics design-expert prompts with fixed sections: design guideline, concern, design method, evaluation method, and knowledge-base reference — the four categories correspond one-to-one by number.
4. Read system-architecture knowledge and locate components: Reads knowledge.md and a base .drawio; evaluates impact per component, identifying change types: modified / affected / new / deleted / reset.
5. Scripted block-diagram updates: Generates changes.json; apply_styles.py copies the base file, updates styles or inserts new components, and saves.
6. Uniform component marking by change type: modified (red border), affected (blue border), new (thick yellow border), deleted (gray transparent dashed), reset (restore default).

Highlights

• Design-assessment guidelines and block diagrams generated in one business flow: One input simultaneously yields a Markdown guideline document and an annotated Drawio file.
• Historical knowledge-driven: Uses historical design records from knowledge.json as background.
• Fine-grained component classification: Examines not only modules but also specific processing units inside Drawio.
• Explicit multilingual delivery rules: Chinese, English, and Japanese template structures stay consistent.

Business Value

• Design-assessment guidelines for a single change point take ~0.5–3 days manually; block-diagram change analysis also takes ~0.5–3 days manually. After automation, processing is expected to drop to minutes, with direct labor cost reduced by 80%+.
• knowledge.json / knowledge.md turns senior architect experience into a retrievable asset.
• Unified number correspondence and fixed component classification reduce manual arrangement errors.
• apply_styles.py automatically modifies XML, reducing manual-editing mistakes.

---