What is heritage-architecture-studio?

Heritage Architecture Studio is a skill for the Hermes AI agent that converts the name of a historic building into a complete architectural generation workflow. You give it something like the Yingxian Wooden Pagoda, and it runs through a multi-step pipeline: analyzing the building's type, style, era, culture, and materials, producing a structured reference prompt, generating a self-contained JSON specification, assembling a multi-view reference board, and finally building an interactive Three.js webpage exhibit that runs in desktop and mobile browsers. The skill identifies architectural features and distinguishes between confirmed facts, inferences, simplifications, and uncertain information. It avoids presenting guesses as precise survey data. It enforces a fixed set of ten standard views, including hero perspective, front, back, left, right, top, bottom, roof detail, and ornament detail, plus at least two additional perspective views. The reference board requires at least eleven separate panels. It adapts its output for standing buildings, roofless ruins, and buildings with restrained decoration. Users can run the full pipeline or limit the scope. You can request only the prompt, only the JSON, the prompt plus JSON, the reference board without the webpage, or the Three.js webpage alone. Supplying reference images adds them as evidence without overriding the requested scope. The skill includes a JSON Schema for validation, a prompt and JSON consistency checker, and fallback behavior that preserves the prompt and JSON when image services fail, continuing to Three.js when feasible. Installation is done through the Hermes skills system with a single command, or by downloading a packaged skill file from the GitHub releases page. The repository contains the skill definition, reference documents, validation scripts, and templates for the JSON schema and the multi-view board prompt. Local validation scripts check repository structure and compile the Python files. The design boundaries are clearly stated. The generated reference boards help lock in silhouette, materials, and readability but do not replace conservation surveying. Structural dimensions prefer authoritative sources, and uncertainty is recorded rather than fabricated. The Three.js output is an interpretable procedural reconstruction, not a museum-grade CAD restoration.