Agentcad Skill Release 0.4.0

This Skill addresses the inability of AI agents to perform 3D modeling directly. In practical development, developers often need AI assistance to design mechanical parts, product prototypes, or architectural models. However, traditional CAD tools require manual operation and cannot be deeply integrated into AI workflows. Agentcad enables AI agents to generate STEP files, PNG renders, mesh exports (STL/GLB/OBJ), and geometric measurement data with a single command by writing build123d or CadQuery Python scripts, all without leaving the command line.

Usage is straightforward: simply describe your 3D design requirements in the conversation, and the AI agent will automatically generate and execute the Python script. The core workflow consists of three steps: first, write the script (no manual imports needed; build123d primitives and tools like show_object are pre-configured); then, run a dry-run to check metrics like volume and dimensions; finally, execute the formal run to output preview images, an interactive HTML viewer, and diff comparison images. Each iteration can be quickly compared using agentcad diff to ensure the design direction is correct.

It is ideal for product managers, agent developers, and backend engineers who need rapid prototyping. Especially for teams already using AI-assisted coding, it seamlessly integrates 3D modeling capabilities into automated workflows, such as generating product showcase images, 3D printing models, or mechanical components. For scenarios requiring frequent parameter adjustments and design validation, Agentcad significantly reduces the time cost of manual modeling.

It is recommended to run agentcad init --name <project_name> to initialize a project before first use and to carefully review agentcad --help for a complete command reference. Note that it primarily relies on Python 3.10-3.12 and requires the agentcad package to be pre-installed. The generated STEP files can be imported into any mainstream CAD software, but the interactive viewer (viewer.html) is only for human users to open—AI agents cannot render HTML, so remember to prompt users to open it themselves.