Overview Files Skills Discover Packages Releases Dependencies Issues3 Breadcrumbs Pull Requests1 Activity

main

John Lauer AI Flow 0.1.54: 3D models checked for which way is up; missing 3D named; the AI told to look 29bd82b 2h ago

AI Flow 0.1.54

Every 3D model is checked for which way is up. A STEP carries no "up" axis, and many are authored Y-up while the board and Fusion are Z-up. KiCad hides this because each footprint carries its model's transform ((model ... (rotate ...) (offset ...))); the Fusion binding used the raw STEP, so on the buck molecule the 47 uF can lay on its flank and the inductor stood on edge. fusion library now applies the footprint's own transform (KiCad's convention), checks every model against its footprint's courtyard, turns a model with no transform upright when its height is clearly not on Z, binds the corrected copy, and names every part it turned. Parts bound under the old rules are bound again on the next fusion library.
The AI is told to look. The library and schematic replies say to check the board in 3D with fusion_take_screenshot (Windows window capture cannot see Fusion's 3D viewport) before calling it done. The skill says why: the AI does not do this on its own.
Parts that lost their 3D are named. fusion schematic counts the board parts whose 3D binding Fusion dropped when they were added live (they show as flat placeholder boxes) and says how to restore it.
Docs: docs/fusion-lane.md, "Which way is up".

Known, next: restoring a dropped 3D binding automatically; the library name clashing with an earlier run of the same project; the blank-clip check flagging a mostly black Fusion board view; pre-board step clips on a Fusion project filming a KiCad window; the schematic's molecule-interface column overlapping the first group.

# AI Flow 0.1.54

- **Every 3D model is checked for which way is up.** A STEP carries no "up" axis, and many are authored Y-up while the board and Fusion are Z-up. KiCad hides this because each footprint carries its model's transform (`(model ... (rotate ...) (offset ...))`); the Fusion binding used the raw STEP, so on the buck molecule the 47 uF can lay on its flank and the inductor stood on edge. `fusion library` now applies the footprint's own transform (KiCad's convention), checks every model against its footprint's courtyard, turns a model with no transform upright when its height is clearly not on Z, binds the corrected copy, and names every part it turned. Parts bound under the old rules are bound again on the next `fusion library`.
- **The AI is told to look.** The library and schematic replies say to check the board in 3D with `fusion_take_screenshot` (Windows window capture cannot see Fusion's 3D viewport) before calling it done. The skill says why: the AI does not do this on its own.
- **Parts that lost their 3D are named.** `fusion schematic` counts the board parts whose 3D binding Fusion dropped when they were added live (they show as flat placeholder boxes) and says how to restore it.
- Docs: `docs/fusion-lane.md`, "Which way is up".

Known, next: restoring a dropped 3D binding automatically; the library name clashing with an earlier run of the same project; the blank-clip check flagging a mostly black Fusion board view; pre-board step clips on a Fusion project filming a KiCad window; the schematic's molecule-interface column overlapping the first group.