LEGO: LLM Skill-Based Platform for Front-End Design Generation

LEGO serves as a cohesive skill-oriented framework for generating front-end designs, breaking down the digital front-end process into six distinct phases. It illustrates agent capabilities through standardized, composable circuit skills within a modular architecture. The platform analyzed more than 100 research papers, chose 11 notable open-source projects, and derived 42 executable circuit skills using a six-step finite state machine approach. The Circuit Skill Builder facilitates skill extraction with linear scalability, while the Agent Skill RAG enables submillisecond retrieval without depending on embedding models. An empirical assessment on a challenging subset of 41 VerilogEval v2 problems, which gpt-5.2-codex struggles to resolve under high reasoning demands, demonstrates the platform's efficacy.