vision-decompose

Decomposes engineering reference images into structured CAD feature trees before Onshape modeling.

Updated Apr 26, 2026
One-click install
npx skills add https://github.com/adityacodepublic/onShape-cad --skill vision-decompose-adityacodepublic
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: vision-decompose
Source: https://github.com/adityacodepublic/onShape-cad/tree/main/skills/vision-decompose
Command: npx skills add https://github.com/adityacodepublic/onShape-cad --skill vision-decompose-adityacodepublic

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When asked to build a CAD part from a reference image, agents often skim the image and misread features (boss vs pocket, complex outlines vs rounded rectangles), causing long, wasteful iteration loops. This Skill enforces a rigorous zoom-and-describe pass that produces a reviewable feature tree before any geometry is created. ## Core Features & Use Cases - Structured Feature Tree Output: Produces a fixed-format decomposition with overview, envelope dimensions, per-feature type/size/position/role, relationships, and explicit uncertainties. - Guided Crop-and-Describe Workflow: Uses load_local_image and crop_image to zoom into each feature and dimension callout at native resolution, reading values like Ø25, R3, and 4X patterns directly from drawings. - User Confirmation Gate: Pauses after decomposition so the user can correct misread features before the build phase begins. - Use Case: A user shares a photo of a motor bracket drawing and asks to model it. The Skill crops into each callout and feature region, then outputs a feature tree (base plate, 4-hole bolt pattern, central boss, fillets) the user confirms before Onshape building starts. ## Quick Start Analyze this engineering drawing image and produce a structured feature decomposition before building the part in Onshape.

Frequently Asked Questions about vision-decompose

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I convert an engineering drawing image into a CAD model?

First decompose the image into a structured feature tree by cropping into each feature and dimension callout at native resolution. Confirm the decomposition with the user, then execute the build phase in Onshape against that verified spec.

How to read dimension callouts from a low-resolution drawing image?

Cache the image with load_local_image, then use crop_image to zoom into each callout region at native resolution. This avoids vision downsampling that clobbers small dimension text like Ø25 or R3.

When should I skip the vision decomposition step?

Skip it when the user has already described the part in plain text rather than providing a reference image. The skill is specifically for image-to-CAD workflows where visual misreading is the main risk.

Why does image-to-CAD modeling misread features like bosses and pockets?

Misreads happen when agents skim the full image instead of cropping into each feature. A cylindrical bump (boss, additive) versus a hole (pocket, subtractive) requires checking polarity, and derived outlines must be compared against other features' silhouettes.

What are the limitations of single-image CAD decomposition?

Single images can hide backside features, ambiguous depths, and unreadable dimensions. The skill handles this by listing explicit uncertainties and asking the user to confirm or correct the feature tree before building.