Pro-Tool Spine · Product/CAD beachhead

Idea → Printable Part

Describe a part. Get a manifold, slicer-ready STL that passes a real printability gate — plus a CAD-interchange STEP and a web-preview GLB. Exact dimensions, restated before render.

The loop

prompt: "wall bracket, 80×40×5mm, four M4 holes, 2mm fillet" → create_cad_model # OpenSCAD → manifold STL + 4-view PNG + printability report → refine_cad_model "+2mm thickness, chamfer top edge" # chain as needed → export_stl # re-validate slicer-readiness → convert to="step" # tessellated-shell STEP for CAD interchange (warns: not parametric) → convert to="glb" # web/AR preview

The steps (artifacts land here on live render)

1 · Spec
dims in mm
80×40×5mm base + 30mm wall · 4× Ø4.5
2 · Render+gate
cad-render
bracket.stl · printable ✓
3 · STEP
FreeCAD cap
fast-follow (OCC)
4 · Preview
format-convert
bracket.glb
Live printability report (cad-validate gate): watertight ✓ · winding-consistent ✓ · positive-volume ✓ · 1 body · 268 verts / 548 faces · volume 25,683 mm³ · bbox 80×40×30 mm · printable: true. STL→GLB preview via the live format-convert cap in 4.5s.
Honesty: the authoring path is OpenSCAD (mesh-only). STL is the lossless print deliverable. convert to="step" yields a tessellated-shell STEP — valid for CAD-tool geometry import, but NOT an editable parametric solid. The spine stamps this warning; true parametric STEP needs a BREP kernel (FreeCAD/OCC cap, fast-follow).

Why it beats raw AI mesh

Every render runs the printability gate (cad-validate): watertight · manifold · positive-volume · bed-fit · min-wall. AI mesh caps (tripo-i3d) usually FAIL it — code-CAD is manufacture-ready.

Recipe

Playbook: public/playbooks/idea-to-printable-part.md · skill: skills/cad-pipeline.md. Registers on /playbooks + /case-studies once the live render lands.